1 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3 * config/rs6000/rs6000.c (rs6000_option_override_internal):
4 Display warning message for -mno-speculate-indirect-jumps.
6 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
8 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
10 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
12 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
14 PR tree-optimization/84225
15 * tree-eh.c (find_trapping_overflow): Only call
16 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
18 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
21 * config/i386/i386.c: Reimplement the check of possible options
22 -mibt/-mshstk conbination. Change error messages.
23 * doc/invoke.texi: Fix a typo: remove extra '='.
25 2018-02-06 Marek Polacek <polacek@redhat.com>
27 PR tree-optimization/84228
28 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
30 2018-02-06 Tamar Christina <tamar.christina@arm.com>
33 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
34 emitted arch directives.
35 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
36 __ARM_FEATURE_COPROC before changing architectures.
38 2018-02-06 Richard Biener <rguenther@suse.de>
40 * config/i386/i386.c (print_reg): Fix typo.
41 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
43 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
45 * configure: Regenerate.
47 2018-02-05 Martin Sebor <msebor@redhat.com>
49 PR tree-optimization/83369
50 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
53 2018-02-05 Martin Liska <mliska@suse.cz>
55 * doc/invoke.texi: Cherry-pick upstream r323995.
57 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
59 * ira.c (ira_init_register_move_cost): Adjust comment.
61 2018-02-05 Martin Liska <mliska@suse.cz>
64 * doc/gcov.texi: Fix typo in documentation.
66 2018-02-05 Martin Liska <mliska@suse.cz>
69 * doc/gcov.texi: Document necessity of --dynamic-list-data when
70 using dlopen functionality.
72 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
74 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
75 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
76 _mm_maskz_range_ss, _mm_mask_range_round_ss,
77 _mm_maskz_range_round_ss): New intrinsics.
78 (__builtin_ia32_rangesd128_round)
79 (__builtin_ia32_rangess128_round): Remove.
80 (__builtin_ia32_rangesd128_mask_round,
81 __builtin_ia32_rangess128_mask_round): New builtins.
82 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
83 __builtin_ia32_rangess128_round): Remove.
84 (__builtin_ia32_rangesd128_mask_round,
85 __builtin_ia32_rangess128_mask_round): New builtins.
86 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
87 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
88 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
89 "<round_saeonly_constraint>")): Changed to ...
90 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
91 "<round_saeonly_scalar_constraint>")): ... this.
92 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
93 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
94 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
95 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
96 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
98 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
100 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
102 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
103 Remove all values except native, 8540 and 8548.
105 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
107 * config/i386/i386.c (ix86_output_function_return): Pass
108 INVALID_REGNUM, instead of -1, as invalid register number to
109 indirect_thunk_name and output_indirect_thunk.
111 2018-02-02 Julia Koval <julia.koval@intel.com>
113 * config.gcc: Add -march=icelake.
114 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
115 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
116 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
117 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
118 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
119 (processor_target_table): Add icelake.
120 (ix86_option_override_internal): Handle new PTAs.
121 (get_builtin_code_for_version): Handle icelake.
122 (M_INTEL_COREI7_ICELAKE): New.
123 (fold_builtin_cpu): Handle icelake.
124 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
125 * doc/invoke.texi: Add -march=icelake.
127 2018-02-02 Julia Koval <julia.koval@intel.com>
129 * config/i386/i386.c (ix86_option_override_internal): Change flags type
131 * wide-int-bitmask.h: New.
133 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
136 * config/i386/i386.md: Replace Pmode with word_mode in
137 builtin_setjmp_setup and builtin_longjmp to support x32.
139 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
143 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
145 (rs6000_supported_cpu_names): New static variable.
146 (linux_cpu_translation_table): Likewise.
147 (elf_platform) <cpu>: Define new static variable and use it.
148 Translate kernel AT_PLATFORM name to canonical name if needed.
149 Error if platform name is unknown.
151 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
154 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
156 2018-02-01 Jeff Law <law@redhat.com>
159 * config/i386/i386.c (release_scratch_register_on_entry): Add new
160 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
161 the scratch if RELEASE_VIA_POP is false.
162 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
163 If we have to save a temporary register, decrement SIZE appropriately.
164 Pass new arguments to release_scratch_register_on_entry.
165 (ix86_adjust_stack_and_probe): Likewise.
166 (ix86_emit_probe_stack_range): Pass new arguments to
167 release_scratch_register_on_entry.
169 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
171 PR rtl-optimization/84157
172 * combine.c (change_zero_ext): Use REG_P predicate in
173 front of HARD_REGISTER_P predicate.
175 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
177 * config/avr/avr.c (avr_option_override): Move disabling of
178 -fdelete-null-pointer-checks to...
179 * common/config/avr/avr-common.c (avr_option_optimization_table):
182 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
184 PR tree-optimization/81635
185 * tree-data-ref.c (split_constant_offset_1): For types that
186 wrap on overflow, try to use range info to prove that wrapping
189 2018-02-01 Renlin Li <renlin.li@arm.com>
192 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
194 (aarch64_register_move_cost): Likewise.
195 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
197 (REG_CLASS_NAMES): Likewise.
198 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
199 TAILCALL_ADDR_REGS. Remove IP registers.
200 * config/aarch64/aarch64.md (Ucs): Update register constraint.
202 2018-02-01 Richard Biener <rguenther@suse.de>
204 * domwalk.h (dom_walker::dom_walker): Add additional constructor
205 for specifying RPO order and allow NULL for that.
206 * domwalk.c (dom_walker::dom_walker): Likewise.
207 (dom_walker::walk): Handle NULL RPO order.
208 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
210 (rewrite_update_dom_walker): Likewise.
211 (mark_def_dom_walker): Likewise.
213 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
215 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
216 (aarch64_maybe_expand_sve_subreg_move): Declare.
217 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
218 * config/aarch64/predicates.md (aarch64_any_register_operand): New
220 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
221 that are semantically a reverse operation.
222 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
223 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
224 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
226 (aarch64_can_change_mode_class): For big-endian, forbid changes
227 between two SVE modes if they have different element sizes.
229 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
231 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
232 the TImode handling for big-endian targets.
234 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
236 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
237 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
239 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
240 Remove BSWAP handing for big-endian targets and use the form of
241 LD1RQ appropariate for the mode.
243 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
245 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
246 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
249 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
252 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
253 check for operands that need to go through aarch64_sve_reload_be.
255 2018-02-01 Jakub Jelinek <jakub@redhat.com>
257 PR tree-optimization/81661
258 PR tree-optimization/84117
259 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
260 * tree-eh.c: Include gimplify.h.
261 (find_trapping_overflow, replace_trapping_overflow,
262 rewrite_to_non_trapping_overflow): New functions.
263 * tree-vect-loop.c: Include tree-eh.h.
264 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
265 * tree-data-ref.c: Include tree-eh.h.
266 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
268 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
270 PR rtl-optimization/84123
271 * combine.c (change_zero_ext): Check if hard register satisfies
272 can_change_dest_mode before calling gen_lowpart_SUBREG.
274 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
277 * ira.c (ira_init_register_move_cost): Remove assert.
279 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
281 PR rtl-optimization/84071
282 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
283 * doc/tm.texi: Regenerate.
285 2018-01-31 Richard Biener <rguenther@suse.de>
287 PR tree-optimization/84132
288 * tree-data-ref.c (analyze_miv_subscript): Properly
289 check whether evolution_function_is_affine_multivariate_p
290 before calling gcd_of_steps_may_divide_p.
292 2018-01-31 Julia Koval <julia.koval@intel.com>
295 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
296 * config/i386/i386.md (rdpid_rex64) New.
297 (rdpid): Make 32bit only.
299 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
302 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
303 an IDENTIFIER_NODE for FUNCTION_TYPE's.
305 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
308 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
310 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
312 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
314 PR rtl-optimization/84071
315 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
316 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
318 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
320 * config/arc/arc.c (arc_handle_aux_attribute): New function.
321 (arc_attribute_table): Add 'aux' attribute.
322 (arc_in_small_data_p): Consider aux like variables.
323 (arc_is_aux_reg_p): New function.
324 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
325 (arc_get_aux_arg): New function.
326 (prepare_move_operands): Handle aux-register access.
327 (arc_handle_aux_attribute): New function.
328 * doc/extend.texi (ARC Variable attributes): Add subsection.
330 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
332 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
333 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
334 (arc_attribute_table): Add 'uncached' attribute.
335 (arc_print_operand): Print '.di' flag for uncached memory
337 (arc_in_small_data_p): Do not consider for small data the uncached
339 (arc_is_uncached_mem_p): New function.
340 * config/arc/predicates.md (compact_store_memory_operand): Check
341 for uncached memory accesses.
342 (nonvol_nonimm_operand): Likewise.
343 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
345 2018-01-31 Jakub Jelinek <jakub@redhat.com>
348 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
349 falign-loops=): Add Optimization flag.
351 2018-01-30 Jeff Law <law@redhat.com>
354 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
355 INT_REGISTERS_SAVED. Check it prior to calling
356 get_scratch_register_on_entry.
357 (ix86_adjust_stack_and_probe): Similarly.
358 (ix86_emit_probe_stack_range): Similarly.
359 (ix86_expand_prologue): Corresponding changes.
361 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
364 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
365 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
367 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
370 * lra-constraints.c (curr_insn_transform): Process AND in the
373 2018-01-30 Jakub Jelinek <jakub@redhat.com>
375 PR rtl-optimization/83986
376 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
377 dependence against last_pending_memory_flush in addition to
380 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
382 PR tree-optimization/81611
383 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
386 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
388 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
391 2018-01-30 Richard Biener <rguenther@suse.de>
392 Jakub Jelinek <jakub@redhat.com>
394 PR tree-optimization/84111
395 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
396 inner loops added during recursion, as they don't have up-to-date
399 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
402 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
403 (can_inline_edge_by_limits_p): ... here.
404 (can_early_inline_edge_p, check_callers,
405 update_caller_keys, update_callee_keys, recursive_inlining,
406 add_new_edges_to_heap, speculation_useful_p,
407 inline_small_functions,
408 inline_small_functions, flatten_function,
409 inline_to_all_callers_1): Update.
411 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
413 * profile-count.c (profile_count::combine_with_ipa_count): Handle
416 2018-01-30 Richard Biener <rguenther@suse.de>
418 PR tree-optimization/83008
419 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
420 invariant and constant vector uses in stmts when they need
423 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
426 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
427 * configure: Regenerate.
429 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
431 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
433 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
434 Use gen_rtx_REG rather than gen_lowpart.
436 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
438 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
439 rather than 0 when creating partial subregs.
441 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
443 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
446 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
449 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
450 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
451 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
452 flags. This restores the settings used before the 2017-07-24.
453 Turning off pre increment/decrement/modify allows IVOPTS to
454 optimize DF/SF loops where the index is an int.
456 2018-01-29 Richard Biener <rguenther@suse.de>
457 Kelvin Nilsen <kelvin@gcc.gnu.org>
460 * tree-vect-stmts.c (vectorizable_call): Don't call
461 targetm.vectorize_builtin_md_vectorized_function if callee is
464 2018-01-22 Carl Love <cel@us.ibm.com>
466 * doc/extend.tex: Fix typo in second arg in
467 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
469 2018-01-29 Richard Biener <rguenther@suse.de>
471 PR tree-optimization/84086
472 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
473 (flush_ssaname_freelist): When SSA names were released reset
476 2018-01-29 Richard Biener <rguenther@suse.de>
478 PR tree-optimization/84057
479 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
480 removed paths when removing edges.
482 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
484 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
485 -mfunction-return=@var{choice}.
487 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
490 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
492 (layout::print_source_line): Likewise.
493 (test_get_line_width_without_trailing_whitespace): Add test cases.
495 2018-01-27 Jakub Jelinek <jakub@redhat.com>
498 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
501 2018-01-26 Jim Wilson <jimw@sifive.com>
503 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
505 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
508 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
510 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
511 and CMP + SUB-immediate -> SUBS.
513 2018-01-26 Martin Sebor <msebor@redhat.com>
515 PR tree-optimization/83896
516 * tree-ssa-strlen.c (get_string_len): Rename...
517 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
518 Avoid assuming length is constant.
519 (handle_char_store): Use HOST_WIDE_INT for string length.
521 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
524 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
525 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
527 2018-01-26 Richard Biener <rguenther@suse.de>
529 PR rtl-optimization/84003
530 * dse.c (record_store): Only record redundant stores when
531 the earlier store aliases at least all accesses the later one does.
533 2018-01-26 Jakub Jelinek <jakub@redhat.com>
535 PR rtl-optimization/83985
536 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
537 REG_CFA_RESTORE insns.
538 (delete_unmarked_insns): Don't ignore separate shrink wrapping
539 REG_CFA_RESTORE insns here.
542 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
543 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
545 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
547 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
548 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
549 (arc_init): Likewise.
550 (arc_override_options): Likewise.
551 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
553 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
555 * config/arc/arc.h (TARGET_DBNZ): Define.
556 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
557 properly set the tune attribute.
558 (dbnz): Use TARGET_DBNZ guard.
559 * config/arc/arc.opt (mtune): Add core3 option.
561 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
563 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
564 recognize new pic like addresses.
565 (arc_delegitimize_address): Clean up.
567 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
569 * config/arc/arc-arches.def: Option mrf16 valid for all
571 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
572 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
573 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
574 * config/arc/arc-tables.opt: Regenerate.
575 * config/arc/arc.c (arc_conditional_register_usage): Handle
576 reduced register file case.
577 (arc_file_start): Set must have build attributes.
578 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
580 * config/arc/arc.opt (mrf16): Add new option.
581 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
582 * config/arc/genmultilib.awk: Handle new mrf16 option.
583 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
584 * config/arc/t-multilib: Regenerate.
585 * doc/invoke.texi (ARC Options): Document mrf16 option.
587 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
589 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
590 * config/arc/arc.c (arc_handle_secure_attribute): New function.
591 (arc_attribute_table): Add 'secure_call' attribute.
592 (arc_print_operand): Print secure call operand.
593 (arc_function_ok_for_sibcall): Don't optimize tail calls when
595 (arc_is_secure_call_p): New function. * config/arc/arc.md
596 (call_i): Add support for sjli instruction.
597 (call_value_i): Likewise.
598 * config/arc/constraints.md (Csc): New constraint.
600 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
601 John Eric Martin <John.Martin@emmicro-us.com>
603 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
604 * config/arc/arc.c (_arc_jli_section): New struct.
605 (arc_jli_section): New type.
606 (rc_jli_sections): New static variable.
607 (arc_handle_jli_attribute): New function.
608 (arc_attribute_table): Add jli_always and jli_fixed attribute.
609 (arc_file_end): New function.
610 (TARGET_ASM_FILE_END): Define.
611 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
612 (arc_add_jli_section): New function.
613 (jli_call_scan): Likewise.
614 (arc_reorg): Call jli_call_scan.
615 (arc_output_addsi): Remove 'S' from printing asm operand.
616 (arc_is_jli_call_p): New function.
617 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
619 (movhi_insn): Likewise.
620 (movsi_insn): Likewise.
621 (movsi_set_cc_insn): Likewise.
622 (loadqi_update): Likewise.
623 (load_zeroextendqisi_update): Likewise.
624 (load_signextendqisi_update): Likewise.
625 (loadhi_update): Likewise.
626 (load_zeroextendhisi_update): Likewise.
627 (load_signextendhisi_update): Likewise.
628 (loadsi_update): Likewise.
629 (loadsf_update): Likewise.
630 (movsicc_insn): Likewise.
631 (bset_insn): Likewise.
632 (bxor_insn): Likewise.
633 (bclr_insn): Likewise.
634 (bmsk_insn): Likewise.
635 (bicsi3_insn): Likewise.
636 (cmpsi_cc_c_insn): Likewise.
637 (movsi_ne): Likewise.
638 (movsi_cond_exec): Likewise.
639 (clrsbsi2): Likewise.
650 (call_i): Remove 'S' asm letter, add jli instruction.
651 (call_value_i): Likewise.
652 * config/arc/arc.op (mjli-always): New option.
653 * config/arc/constraints.md (Cji): New constraint.
654 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
656 (subsf3_fpx): Likewise.
657 (mulsf3_fpx): Likewise.
658 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
660 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
662 * doc/invoke.texi (ARC): Document mjli-always option.
664 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
666 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
667 and use incw and decw where possible
668 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
670 2018-01-26 Richard Biener <rguenther@suse.de>
672 PR tree-optimization/81082
673 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
674 association if it requires casting to unsigned.
675 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
676 from fold_plusminus_mult_expr to catch important cases late when
677 range info is available.
679 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
681 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
682 * configure.ac (hidden_linkonce): New test.
683 * configure: Regenerate.
684 * config.in: Regenerate.
686 2018-01-26 Julia Koval <julia.koval@intel.com>
688 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
689 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
690 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
691 _mm_mask_bitshuffle_epi64_mask): Fix type.
692 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
693 USI_FTYPE_V4DI_V4DI_USI): Remove.
694 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
695 __builtin_ia32_vpshufbitqmb256_mask,
696 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
697 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
698 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
700 2018-01-26 Alan Modra <amodra@gmail.com>
703 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
704 UNSPEC_VBPERMQ. Sort other unspecs.
706 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
708 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
710 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
713 * predict.c (drop_profile): Do not push/pop cfun; update also
715 (handle_missing_profiles): Fix logic looking for zero profiles.
717 2018-01-25 Jakub Jelinek <jakub@redhat.com>
720 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
721 on functions with #pragma omp declare simd or functions with simd
723 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
724 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
725 Remove trailing \n from warning_at calls.
727 2018-01-25 Tom de Vries <tom@codesourcery.com>
730 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
731 for neutered workers.
733 2018-01-24 Joseph Myers <joseph@codesourcery.com>
736 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
737 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
739 2018-01-24 Jeff Law <law@redhat.com>
742 * i386.c (get_probe_interval): Move to earlier point.
743 (ix86_compute_frame_layout): If -fstack-clash-protection and
744 the frame is larger than the probe interval, then use pushes
745 to save registers rather than reg->mem moves.
746 (ix86_expand_prologue): Remove conditional for int_registers_saved
749 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
752 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
753 min/max for never referenced object.
755 2018-01-24 Jakub Jelinek <jakub@redhat.com>
758 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
760 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
761 attributes from DECL_ATTRIBUTES (decl) without affecting
762 DECL_ATTRIBUTES (current_function_decl).
763 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
764 functions with non-NULL DECL_ABSTRACT_ORIGIN.
766 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
768 PR tree-optimization/83979
769 * fold-const.c (fold_comparison): Use constant_boolean_node
770 instead of boolean_{true,false}_node.
772 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
774 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
777 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
779 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
780 Simplify the clause that sets the length attribute.
781 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
782 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
783 clause that sets the length attribute.
784 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
786 2018-01-24 Tom de Vries <tom@codesourcery.com>
789 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
790 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
791 Add strict parameter.
792 (prevent_branch_around_nothing): Insert dummy insn between branch to
793 label and label with no ptx insn inbetween.
794 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
796 2018-01-24 Tom de Vries <tom@codesourcery.com>
799 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
800 for neutered threads in warp.
801 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
803 2018-01-24 Richard Biener <rguenther@suse.de>
805 PR tree-optimization/83176
806 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
809 2018-01-24 Richard Biener <rguenther@suse.de>
811 PR tree-optimization/82819
812 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
813 code generating pluses that are no-ops in the target precision.
815 2018-01-24 Richard Biener <rguenther@suse.de>
818 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
820 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
822 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
823 to merge probabilities.
824 * predict.c (probably_never_executed): Also mark as cold functions
825 with global 0 profile and guessed local profile.
826 * profile-count.c (profile_probability::combine_with_count): New
828 * profile-count.h (profile_probability::operator*,
829 profile_probability::operator*=, profile_probability::operator/,
830 profile_probability::operator/=): Reduce precision to adjusted
831 and set value to guessed on contradictory divisions.
832 (profile_probability::combine_with_freq): Remove.
833 (profile_probability::combine_wiht_count): Declare.
834 (profile_count::force_nonzero):: Set to adjusted.
835 (profile_count::probability_in):: Set quality to adjusted.
836 * tree-ssa-tail-merge.c (replace_block_by): Use
839 2018-01-23 Andrew Waterman <andrew@sifive.com>
840 Jim Wilson <jimw@sifive.com>
842 * config/riscv/riscv.c (riscv_stack_boundary): New.
843 (riscv_option_override): Set riscv_stack_boundary. Handle
844 riscv_preferred_stack_boundary_arg.
845 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
846 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
847 (STACK_BOUNDARY): Set to riscv_stack_boundary.
848 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
849 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
850 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
852 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
855 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
856 of struct ix86_frame.
857 (ix86_expand_epilogue): Likewise. Add a local variable for
858 the reg_save_offset field in struct ix86_frame.
860 2018-01-23 Bin Cheng <bin.cheng@arm.com>
862 PR tree-optimization/82604
863 * tree-loop-distribution.c (enum partition_kind): New enum item
864 PKIND_PARTIAL_MEMSET.
865 (partition_builtin_p): Support above new enum item.
866 (generate_code_for_partition): Ditto.
867 (compute_access_range): Differentiate cases that equality can be
868 proven at all loops, the innermost loops or no loops.
869 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
870 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
871 (finalize_partitions, distribute_loop): Don't fuse partition of
872 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
873 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
876 2018-01-23 Martin Liska <mliska@suse.cz>
878 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
879 order to ignore the predictor.
880 (PRED_POLYMORPHIC_CALL): Likewise.
881 (PRED_RECURSIVE_CALL): Likewise.
883 2018-01-23 Martin Liska <mliska@suse.cz>
885 * tree-profile.c (tree_profiling): Print function header to
886 aware reader which function we are working on.
887 * value-prof.c (gimple_find_values_to_profile): Do not print
888 not interesting value histograms.
890 2018-01-23 Martin Liska <mliska@suse.cz>
892 * profile-count.h (enum profile_quality): Add
893 profile_uninitialized as the first value. Do not number values
894 as they are zero based.
895 (profile_count::verify): Update sanity check.
896 (profile_probability::verify): Likewise.
898 2018-01-23 Nathan Sidwell <nathan@acm.org>
900 * doc/invoke.texi (ffor-scope): Deprecate.
902 2018-01-23 David Malcolm <dmalcolm@redhat.com>
904 PR tree-optimization/83510
905 * domwalk.c (set_all_edges_as_executable): New function.
906 (dom_walker::dom_walker): Convert bool param
907 "skip_unreachable_blocks" to enum reachability. Move setup of
908 edge flags to set_all_edges_as_executable and only do it when
909 reachability is REACHABLE_BLOCKS.
910 * domwalk.h (enum dom_walker::reachability): New enum.
911 (dom_walker::dom_walker): Convert bool param
912 "skip_unreachable_blocks" to enum reachability.
913 (set_all_edges_as_executable): New decl.
914 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
915 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
917 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
918 but converting true to REACHABLE_BLOCKS.
919 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
921 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
922 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
923 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
925 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
926 if check_all_array_refs will be called.
928 2018-01-23 David Malcolm <dmalcolm@redhat.com>
930 * tree.c (selftest::test_location_wrappers): Add more test
933 2018-01-23 David Malcolm <dmalcolm@redhat.com>
935 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
936 (selftest::test_bit_in_range): Likewise.
938 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
941 * doc/sourcebuild.texi (vect_float): Say that the selector
942 only describes the situation when -funsafe-math-optimizations is on.
943 (vect_float_strict): Document.
945 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
947 PR tree-optimization/83965
948 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
949 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
950 instead of checking only for a reduction.
951 (vect_recog_widen_sum_pattern): Likewise.
953 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
955 * predict.c (probably_never_executed): Only use precise profile info.
956 (compute_function_frequency): Skip after inlining hack since we now
957 have quality checking.
959 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
961 * profile-count.h (profile_probability::very_unlikely,
962 profile_probability::unlikely, profile_probability::even): Set
963 precision to guessed.
965 2018-01-23 Richard Biener <rguenther@suse.de>
967 PR tree-optimization/83963
968 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
969 Properly terminate dominator walk when crossing the exit edge not
970 when visiting its source block.
972 2018-01-23 Jakub Jelinek <jakub@redhat.com>
975 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
976 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
978 2018-01-22 Jakub Jelinek <jakub@redhat.com>
980 PR tree-optimization/83957
981 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
982 semicolon after for body surrounded by braces.
984 PR tree-optimization/83081
985 * profile-count.h (profile_probability::split): New method.
986 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
987 Use profile_probability::split.
988 (do_compare_rtx_and_jump): Fix adjustment of probabilities
989 when splitting a single conditional jump into 2.
991 2018-01-22 David Malcolm <dmalcolm@redhat.com>
993 PR tree-optimization/69452
994 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
997 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
999 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1000 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1001 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1003 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1005 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1006 * config/rl78/rl78.md: New define_expand "movdi"
1007 * config/rl78/rl78.c: New function definition rl78_split_movdi
1009 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1012 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1014 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1015 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1016 128-bit to produce an UNSPEC move to get the double word with the
1017 signbit and then a shift directly to do signbit.
1018 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1019 implementation with a new version that just does either a direct
1020 move or a regular move. Move memory interface to separate insns.
1021 Move insns so they are next to the expander.
1022 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1023 with signbit move. Split big and little endian case.
1024 (signbit<mode>2_dm_mem_le): Likewise.
1025 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1026 (signbit<mode>2_dm2): Likewise.
1028 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1030 * config/rl78/rl78.md: New define_expand "anddi3".
1032 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1034 * config/rl78/rl78.md: New define_expand "umindi3".
1036 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1038 * config/rl78/rl78.md: New define_expand "smindi3".
1040 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1042 * config/rl78/rl78.md: New define_expand "smaxdi3".
1044 2018-01-22 Carl Love <cel@us.ibm.com>
1046 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1047 LVX_V1TI): Add macro expansion.
1048 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1049 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1050 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1051 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1052 Change check to determine if the instruction is a byte reversing
1053 entry. Fix typo in comment.
1054 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1055 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1056 Add def_builtin calls for new builtins.
1057 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1058 Add define_insn expansion.
1060 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1062 * config/rl78/rl78.md: New define_expand "umaxdi3".
1064 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1066 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1067 for non-QImode registers
1069 2018-01-22 Richard Biener <rguenther@suse.de>
1071 PR tree-optimization/83963
1072 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1073 including the loop exit block.
1074 (scop_detection::merge_sese): Likewise.
1075 (scop_detection::add_scop): Do it here instead.
1077 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1079 * doc/sourcebuild.texi (arm_softfloat): Document.
1081 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1084 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1085 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1086 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1088 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1089 David Edelsohn <dje.gcc@gmail.com>
1092 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1093 Change "crset eq" to "crset 2".
1094 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1095 (*call_indirect_aix<mode>_nospec): Likewise.
1096 (*call_value_indirect_aix<mode>_nospec): Likewise.
1097 (*call_indirect_elfv2<mode>_nospec): Likewise.
1098 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1099 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1100 change assembly output from . to $.
1101 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1102 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1103 (*tablejump<mode>_internal1_nospec): Likewise.
1105 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1108 * config/sh/sh_optimize_sett_clrt.cc:
1109 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1111 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1113 PR tree-optimization/83940
1114 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1115 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1116 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1117 use it to pass back the definition type.
1118 (vect_check_store_rhs): Likewise rhs_dt_out.
1119 (vect_build_gather_load_calls): Add a mask_dt argument and use
1120 it instead of a call to vect_is_simple_use.
1121 (vectorizable_store): Update calls to vect_check_load_store_mask
1122 and vect_check_store_rhs. Use the dt returned by the latter instead
1123 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1124 instead of calls to vect_is_simple_use. Pass the scalar rather
1125 than the vector operand to vect_is_simple_use when handling
1126 second and subsequent copies of an rhs value.
1127 (vectorizable_load): Update calls to vect_check_load_store_mask
1128 and vect_build_gather_load_calls. Use the cached mask_dt and
1129 gs_info.offset_dt instead of calls to vect_is_simple_use.
1131 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1134 * tree-emutls.c: Include gimplify.h.
1135 (lower_emutls_2): New function.
1136 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1137 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1138 it before further processing.
1141 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1142 UINTVAL (trueop1) instead of INTVAL (op1).
1144 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1148 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1149 INCOMING_FRAME_SP_OFFSET if not defined.
1150 (scan_trace): Add ENTRY argument. If true and
1151 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1152 emit a note to adjust the CFA offset.
1153 (create_cfi_notes): Adjust scan_trace callers.
1154 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1155 INCOMING_FRAME_SP_OFFSET in the CIE.
1156 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1157 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1159 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1160 * doc/tm.texi: Regenerated.
1162 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1164 PR rtl-optimization/83147
1165 * lra-constraints.c (remove_inheritance_pseudos): Use
1166 lra_substitute_pseudo_within_insn.
1168 2018-01-19 Tom de Vries <tom@codesourcery.com>
1169 Cesar Philippidis <cesar@codesourcery.com>
1172 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1174 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1177 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1178 spaces for function labels.
1180 2018-01-19 Martin Liska <mliska@suse.cz>
1182 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1183 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1184 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1185 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1186 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1187 (PRED_CONST_RETURN): Change from 69 to 65.
1188 (PRED_NULL_RETURN): Change from 91 to 71.
1189 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1190 (PRED_LOOP_GUARD): Change from 66 to 73.
1192 2018-01-19 Martin Liska <mliska@suse.cz>
1194 * predict.c (predict_insn_def): Add new assert.
1195 (struct branch_predictor): Change type to signed integer.
1196 (test_prediction_value_range): Amend test to cover
1198 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1199 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1200 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1201 (PRED_LOOP_IV_COMPARE): Likewise.
1202 * predict.h (PROB_UNINITIALIZED): Define new constant.
1204 2018-01-19 Martin Liska <mliska@suse.cz>
1206 * predict.c (dump_prediction): Add new format for
1207 analyze_brprob.py script which is enabled with -details
1209 * profile-count.h (precise_p): New function.
1211 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1213 PR tree-optimization/83922
1214 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1215 there are no statements that need masking.
1216 (vect_active_double_reduction_p): New function.
1217 (vect_analyze_loop_operations): Use it when handling phis that
1218 are not in the loop header.
1220 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1222 PR tree-optimization/83914
1223 * tree-vect-loop.c (vectorizable_induction): Don't convert
1224 init_expr or apply the peeling adjustment for inductions
1225 that are nested within the vectorized loop.
1227 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1229 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1232 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1236 * function.h (gimplify_parameters): Add gimple_seq * argument.
1237 * function.c: Include gimple.h and options.h.
1238 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1239 for the added local temporaries if needed.
1240 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1241 if there are any parameter cleanups, wrap whole body into a
1242 try/finally with the cleanups.
1244 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1247 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1248 Use GET_MODE_CLASS for scalar floating point.
1250 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1254 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1255 Fix call of call_cgraph_insertion_hooks.
1257 2018-01-18 Martin Sebor <msebor@redhat.com>
1259 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1261 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1264 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1267 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1270 * common.opt: (-ffile-prefix-map): New option.
1271 * opts.c (common_handle_option): Defer it.
1272 * opts-global.c (handle_common_deferred_options): Handle it.
1273 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1274 * file-prefix-map.h: New file.
1275 (remap_debug_filename, add_debug_prefix_map): ...here.
1276 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1277 * final.c (debug_prefix_map, add_debug_prefix_map
1278 remap_debug_filename): Move to...
1279 * file-prefix-map.c: New file.
1280 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1281 generalize, get rid of alloca(), use strrchr() instead of strchr().
1282 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1283 Implement in terms of add_prefix_map().
1284 (remap_macro_filename, remap_debug_filename): Implement in term of
1286 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1287 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1288 * dbxout.c: Include file-prefix-map.h.
1289 * varasm.c: Likewise.
1290 * vmsdbgout.c: Likewise.
1291 * xcoffout.c: Likewise.
1292 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1293 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1294 * doc/invoke.texi (-ffile-prefix-map): Document.
1295 (-fdebug-prefix-map): Update description.
1297 2018-01-18 Martin Liska <mliska@suse.cz>
1299 * config/i386/i386.c (indirect_thunk_name): Document that also
1301 (output_indirect_thunk): Document why both instructions
1302 (pause and lfence) are generated.
1304 2018-01-18 Richard Biener <rguenther@suse.de>
1306 PR tree-optimization/83887
1307 * graphite-scop-detection.c
1308 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1309 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1310 (scop_detection::merge_sese): Re-implement with a flood-fill
1311 algorithm that properly finds a SESE region if it exists.
1313 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1316 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1317 pointer_diff optimizations use view_convert instead of convert.
1319 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1321 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1322 Generate different code for -mno-speculate-indirect-jumps.
1323 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1324 (*call_indirect_aix<mode>): Disable for
1325 -mno-speculate-indirect-jumps.
1326 (*call_indirect_aix<mode>_nospec): New define_insn.
1327 (*call_value_indirect_aix<mode>): Disable for
1328 -mno-speculate-indirect-jumps.
1329 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1330 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1331 -mno-speculate-indirect-jumps.
1332 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1334 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1336 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1337 long double type, set the flags for noting the default long double
1338 type, even if we don't pass or return a long double type.
1340 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1343 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1346 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1348 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1349 support for merge[hl].
1350 (fold_mergehl_helper): New helper function.
1351 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1352 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1353 (altivec_vmrglw_direct): Add xxmrglw insn.
1355 2018-01-17 Andrew Waterman <andrew@sifive.com>
1357 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1358 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1360 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1363 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1364 call the lto_location_cache before reading the
1365 DECL_SOURCE_LOCATION of the types.
1367 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1368 Richard Sandiford <richard.sandiford@linaro.org>
1370 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1371 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1372 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1373 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1374 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1376 * config/aarch64/constraints.md (aarch64_movti_operand):
1378 * config/aarch64/predicates.md (Uti): Add new constraint.
1380 2018-01-17 Carl Love <cel@us.ibm.com>
1381 * config/rs6000/vsx.md (define_expand xl_len_r,
1382 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1383 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1385 (define_expand, define_insn): Move the shift left from the
1386 define_insn to the define_expand for lxvl and stxvl instructions.
1387 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1388 and XL_LEN_R definitions to PURE.
1390 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1392 * config/i386/i386.c (indirect_thunk_name): Declare regno
1393 as unsigned int. Compare regno with INVALID_REGNUM.
1394 (output_indirect_thunk): Ditto.
1395 (output_indirect_thunk_function): Ditto.
1396 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1397 in the call to output_indirect_thunk_function.
1399 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1402 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1403 rather than the size of inner_type to determine the stack slot size
1404 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1406 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1409 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1412 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1414 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1415 endian Linux systems to optionally enable multilibs for selecting
1416 the long double type if the user configured an explicit type.
1417 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1418 have no long double multilibs if not defined.
1419 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1420 warn if the user used -mabi={ieee,ibm}longdouble and we built
1421 multilibs for long double.
1422 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1423 appropriate multilib option.
1424 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1426 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1427 for building long double multilibs.
1428 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1430 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1432 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1435 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1437 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1440 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1443 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1446 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1448 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1449 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1451 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1453 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1454 different rtl trees depending on TARGET_64BIT.
1455 (rs6000_gen_lvx): Likewise.
1457 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1459 * config/visium/visium.md (nop): Tweak comment.
1460 (hazard_nop): Likewise.
1462 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1464 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1465 -mspeculate-indirect-jumps.
1466 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1467 for -mno-speculate-indirect-jumps.
1468 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1469 (*call_value_indirect_elfv2<mode>): Disable for
1470 -mno-speculate-indirect-jumps.
1471 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1472 (indirect_jump): Emit different RTL for
1473 -mno-speculate-indirect-jumps.
1474 (*indirect_jump<mode>): Disable for
1475 -mno-speculate-indirect-jumps.
1476 (*indirect_jump<mode>_nospec): New define_insn.
1477 (tablejump): Emit different RTL for
1478 -mno-speculate-indirect-jumps.
1479 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1480 (tablejumpsi_nospec): New define_expand.
1481 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1482 (tablejumpdi_nospec): New define_expand.
1483 (*tablejump<mode>_internal1): Disable for
1484 -mno-speculate-indirect-jumps.
1485 (*tablejump<mode>_internal1_nospec): New define_insn.
1486 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1489 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1491 * caller-save.c (insert_save): Drop unnecessary parameter. All
1494 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1495 Richard Biener <rguenth@suse.de>
1498 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1499 return early, inline manually is_gimple_sizepos. Make sure if we
1500 call gimplify_expr we don't end up with a gimple constant.
1501 * tree.c (variably_modified_type_p): Don't return true for
1502 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1503 * gimplify.h (is_gimple_sizepos): Remove.
1505 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1507 PR tree-optimization/83857
1508 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1509 vectorizable_live_operation for pure SLP statements.
1510 (vectorizable_live_operation): Handle PHIs.
1512 2018-01-16 Richard Biener <rguenther@suse.de>
1514 PR tree-optimization/83867
1515 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1516 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1518 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1521 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1522 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1523 If off is not INTEGER_CST, issue a may not be aligned warning
1524 rather than isn't aligned. Use isn%'t rather than isn't.
1525 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1527 <case MULT_EXPR>: Improve the case when bottom and one of the
1528 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1529 operand, in that case check if the other operand is multiple of
1530 bottom divided by the INTEGER_CST operand.
1532 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1535 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1536 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1537 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1538 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1539 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1540 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1541 (pa_function_arg_size): New function.
1543 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1545 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1546 in a separate statement.
1548 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1550 PR tree-optimization/83847
1551 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1552 group gathers and scatters.
1554 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1556 PR rtl-optimization/86620
1557 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1559 PR rtl-optimization/83213
1560 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1561 to last if both are JUMP_INSNs.
1563 PR tree-optimization/83843
1564 * gimple-ssa-store-merging.c
1565 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1566 store_immediate_info for bswap/nop orig_stores.
1568 2018-01-15 Andrew Waterman <andrew@sifive.com>
1570 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1572 <UDIV>: Increase cost if !TARGET_DIV.
1574 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1576 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1577 (define_attr "cr_logical_3op"): New.
1578 (cceq_ior_compare): Adjust.
1579 (cceq_ior_compare_complement): Adjust.
1580 (*cceq_rev_compare): Adjust.
1581 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1582 (is_cracked_insn): Adjust.
1583 (insn_must_be_first_in_group): Adjust.
1584 * config/rs6000/40x.md: Adjust.
1585 * config/rs6000/440.md: Adjust.
1586 * config/rs6000/476.md: Adjust.
1587 * config/rs6000/601.md: Adjust.
1588 * config/rs6000/603.md: Adjust.
1589 * config/rs6000/6xx.md: Adjust.
1590 * config/rs6000/7450.md: Adjust.
1591 * config/rs6000/7xx.md: Adjust.
1592 * config/rs6000/8540.md: Adjust.
1593 * config/rs6000/cell.md: Adjust.
1594 * config/rs6000/e300c2c3.md: Adjust.
1595 * config/rs6000/e500mc.md: Adjust.
1596 * config/rs6000/e500mc64.md: Adjust.
1597 * config/rs6000/e5500.md: Adjust.
1598 * config/rs6000/e6500.md: Adjust.
1599 * config/rs6000/mpc.md: Adjust.
1600 * config/rs6000/power4.md: Adjust.
1601 * config/rs6000/power5.md: Adjust.
1602 * config/rs6000/power6.md: Adjust.
1603 * config/rs6000/power7.md: Adjust.
1604 * config/rs6000/power8.md: Adjust.
1605 * config/rs6000/power9.md: Adjust.
1606 * config/rs6000/rs64.md: Adjust.
1607 * config/rs6000/titan.md: Adjust.
1609 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1611 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1612 ix86_indirect_branch_register logic.
1614 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1616 * config/i386/constraints.md (Bs): Update
1617 ix86_indirect_branch_register check. Don't check
1618 ix86_indirect_branch_register with GOT_memory_operand.
1620 * config/i386/predicates.md (GOT_memory_operand): Don't check
1621 ix86_indirect_branch_register here.
1622 (GOT32_symbol_operand): Likewise.
1624 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1626 * config/i386/predicates.md (constant_call_address_operand):
1627 Rewrite ix86_indirect_branch_register logic.
1628 (sibcall_insn_operand): Likewise.
1630 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1632 * config/i386/constraints.md (Bs): Replace
1633 ix86_indirect_branch_thunk_register with
1634 ix86_indirect_branch_register.
1636 * config/i386/i386.md (indirect_jump): Likewise.
1637 (tablejump): Likewise.
1638 (*sibcall_memory): Likewise.
1639 (*sibcall_value_memory): Likewise.
1640 Peepholes of indirect call and jump via memory: Likewise.
1641 * config/i386/i386.opt: Likewise.
1642 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1643 (GOT_memory_operand): Likewise.
1644 (call_insn_operand): Likewise.
1645 (sibcall_insn_operand): Likewise.
1646 (GOT32_symbol_operand): Likewise.
1648 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1651 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1652 type rather than type addr's type points to.
1653 (expand_omp_atomic_mutex): Likewise.
1654 (expand_omp_atomic): Likewise.
1656 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1659 * config/i386/i386.c (output_indirect_thunk_function): Use
1660 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1661 for __x86_return_thunk.
1663 2018-01-15 Richard Biener <rguenther@suse.de>
1666 * expmed.c (extract_bit_field_1): Fix typo.
1668 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1671 * config/arm/iterators.md (VF): New mode iterator.
1672 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1673 Remove integer-related logic from pattern.
1674 (neon_vabd<mode>_3): Likewise.
1676 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1679 * common.opt (fstrict-overflow): No longer an alias.
1680 (fwrapv-pointer): New option.
1681 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1682 also for pointer types based on flag_wrapv_pointer.
1683 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1684 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1685 opts->x_flag_wrapv got set.
1686 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1687 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1688 POINTER_TYPE_OVERFLOW_UNDEFINED.
1689 * match.pd: Likewise in address comparison pattern.
1690 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1692 2018-01-15 Richard Biener <rguenther@suse.de>
1695 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1696 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1697 Reset type names to their identifier if their TYPE_DECL doesn't
1698 have linkage (and thus is used for ODR and devirt).
1699 (save_debug_info_for_decl): Remove.
1700 (save_debug_info_for_type): Likewise.
1701 (add_tree_to_fld_list): Adjust.
1702 * tree-pretty-print.c (dump_generic_node): Make dumping of
1703 type names more robust.
1705 2018-01-15 Richard Biener <rguenther@suse.de>
1707 * BASE-VER: Bump to 8.0.1.
1709 2018-01-14 Martin Sebor <msebor@redhat.com>
1712 * builtins.c (check_access): Avoid warning when the no-warning bit
1715 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1717 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1718 * ira-color (allocno_hard_regs_compare): Likewise.
1720 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1723 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1724 Use .pushsection/.popsection.
1726 2018-01-14 Martin Sebor <msebor@redhat.com>
1729 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1731 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1733 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1734 entry from extra_headers.
1735 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1736 extra_headers, make the list bitwise identical to the i?86-*-* one.
1738 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1740 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1741 -mcmodel=large with -mindirect-branch=thunk,
1742 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1743 -mfunction-return=thunk-extern.
1744 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1745 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1746 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1748 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1750 * config/i386/i386.c (print_reg): Print the name of the full
1751 integer register without '%'.
1752 (ix86_print_operand): Handle 'V'.
1753 * doc/extend.texi: Document 'V' modifier.
1755 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1757 * config/i386/constraints.md (Bs): Disallow memory operand for
1758 -mindirect-branch-register.
1760 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1761 (GOT_memory_operand): Likewise.
1762 (call_insn_operand): Likewise.
1763 (sibcall_insn_operand): Likewise.
1764 (GOT32_symbol_operand): Likewise.
1765 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1766 for -mindirect-branch-register.
1767 (tablejump): Likewise.
1768 (*sibcall_memory): Likewise.
1769 (*sibcall_value_memory): Likewise.
1770 Disallow peepholes of indirect call and jump via memory for
1771 -mindirect-branch-register.
1772 (*call_pop): Replace m with Bw.
1773 (*call_value_pop): Likewise.
1774 (*sibcall_pop_memory): Replace m with Bs.
1775 * config/i386/i386.opt (mindirect-branch-register): New option.
1776 * doc/invoke.texi: Document -mindirect-branch-register option.
1778 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1780 * config/i386/i386-protos.h (ix86_output_function_return): New.
1781 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1782 set function_return_type.
1783 (indirect_thunk_name): Add ret_p to indicate thunk for function
1785 (output_indirect_thunk_function): Pass false to
1786 indirect_thunk_name.
1787 (ix86_output_indirect_branch_via_reg): Likewise.
1788 (ix86_output_indirect_branch_via_push): Likewise.
1789 (output_indirect_thunk_function): Create alias for function
1790 return thunk if regno < 0.
1791 (ix86_output_function_return): New function.
1792 (ix86_handle_fndecl_attribute): Handle function_return.
1793 (ix86_attribute_table): Add function_return.
1794 * config/i386/i386.h (machine_function): Add
1795 function_return_type.
1796 * config/i386/i386.md (simple_return_internal): Use
1797 ix86_output_function_return.
1798 (simple_return_internal_long): Likewise.
1799 * config/i386/i386.opt (mfunction-return=): New option.
1800 (indirect_branch): Mention -mfunction-return=.
1801 * doc/extend.texi: Document function_return function attribute.
1802 * doc/invoke.texi: Document -mfunction-return= option.
1804 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1806 * config/i386/i386-opts.h (indirect_branch): New.
1807 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1808 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1809 with local indirect jump when converting indirect call and jump.
1810 (ix86_set_indirect_branch_type): New.
1811 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1812 (indirectlabelno): New.
1813 (indirect_thunk_needed): Likewise.
1814 (indirect_thunk_bnd_needed): Likewise.
1815 (indirect_thunks_used): Likewise.
1816 (indirect_thunks_bnd_used): Likewise.
1817 (INDIRECT_LABEL): Likewise.
1818 (indirect_thunk_name): Likewise.
1819 (output_indirect_thunk): Likewise.
1820 (output_indirect_thunk_function): Likewise.
1821 (ix86_output_indirect_branch_via_reg): Likewise.
1822 (ix86_output_indirect_branch_via_push): Likewise.
1823 (ix86_output_indirect_branch): Likewise.
1824 (ix86_output_indirect_jmp): Likewise.
1825 (ix86_code_end): Call output_indirect_thunk_function if needed.
1826 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1828 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1829 (ix86_attribute_table): Add indirect_branch.
1830 * config/i386/i386.h (machine_function): Add indirect_branch_type
1831 and has_local_indirect_jump.
1832 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1834 (tablejump): Likewise.
1835 (*indirect_jump): Use ix86_output_indirect_jmp.
1836 (*tablejump_1): Likewise.
1837 (simple_return_indirect_internal): Likewise.
1838 * config/i386/i386.opt (mindirect-branch=): New option.
1839 (indirect_branch): New.
1842 (thunk-inline): Likewise.
1843 (thunk-extern): Likewise.
1844 * doc/extend.texi: Document indirect_branch function attribute.
1845 * doc/invoke.texi: Document -mindirect-branch= option.
1847 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1850 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1852 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1854 * ipa-inline.c (want_inline_small_function_p): Return false if
1855 inlining has already failed with CIF_FINAL_ERROR.
1856 (update_caller_keys): Call want_inline_small_function_p before
1858 (update_callee_keys): Likewise.
1860 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1862 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1864 (rs6000_quadword_masked_address_p): Likewise.
1865 (quad_aligned_load_p): Likewise.
1866 (quad_aligned_store_p): Likewise.
1867 (const_load_sequence_p): Add comment to describe the outer-most loop.
1868 (mimic_memory_attributes_and_flags): New function.
1869 (rs6000_gen_stvx): Likewise.
1870 (replace_swapped_aligned_store): Likewise.
1871 (rs6000_gen_lvx): Likewise.
1872 (replace_swapped_aligned_load): Likewise.
1873 (replace_swapped_load_constant): Capitalize argument name in
1874 comment describing this function.
1875 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1876 and stores that access quad-word aligned addresses and replace
1877 with stvx or lvx instructions when appropriate.
1878 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1879 New function prototype.
1880 (rs6000_quadword_masked_address_p): Likewise.
1881 (rs6000_gen_lvx): Likewise.
1882 (rs6000_gen_stvx): Likewise.
1883 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1884 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1885 when memory address is aligned.
1886 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1887 this split to select lvx instruction when memory address is aligned.
1888 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1889 instruction when memory address is aligned.
1890 (*vsx_le_perm_load_v16qi): Likewise.
1891 (four unnamed splitters): Modify to select the stvx instruction
1892 when memory is aligned.
1894 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1896 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1897 which appears in the queue multiple times.
1899 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1900 Alan Hayward <alan.hayward@arm.com>
1901 David Sherwood <david.sherwood@arm.com>
1903 * tree-vectorizer.h (vec_lower_bound): New structure.
1904 (_loop_vec_info): Add check_nonzero and lower_bounds.
1905 (LOOP_VINFO_CHECK_NONZERO): New macro.
1906 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1907 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1908 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1909 fields. Make seg_len the distance travelled, not including the
1911 (dr_direction_indicator): Declare.
1912 (dr_zero_step_indicator): Likewise.
1913 (dr_known_forward_stride_p): Likewise.
1914 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1916 (runtime_alias_check_p): Allow runtime alias checks with
1918 (operator ==): Compare access_size and align.
1919 (prune_runtime_alias_test_list): Rework for new distinction between
1920 the access_size and seg_len.
1921 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1923 (get_segment_min_max): New function.
1924 (create_intersect_range_checks): Use it.
1925 (dr_step_indicator): New function.
1926 (dr_direction_indicator): Likewise.
1927 (dr_zero_step_indicator): Likewise.
1928 (dr_known_forward_stride_p): Likewise.
1929 * tree-loop-distribution.c (data_ref_segment_size): Return
1930 DR_STEP * (niters - 1).
1931 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1933 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1934 (vect_preserves_scalar_order_p): New function, split out from...
1935 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1936 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1937 (vect_vfa_access_size): New function.
1938 (vect_vfa_align): Likewise.
1939 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1940 (dump_lower_bound): New function.
1941 (vect_check_lower_bound): Likewise.
1942 (vect_small_gap_p): Likewise.
1943 (vectorizable_with_step_bound_p): Likewise.
1944 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1945 depencies if the vectorization factor is 1. Convert the checks
1946 for nonzero steps into checks on the bounds of DR_STEP. Try using
1947 a bunds check for variable steps if the minimum required step is
1948 relatively small. Update calls to the dr_with_seg_len
1949 constructor and to vect_compile_time_alias.
1950 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1952 (vect_loop_versioning): Call it.
1953 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1955 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1957 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1958 Alan Hayward <alan.hayward@arm.com>
1959 David Sherwood <david.sherwood@arm.com>
1961 * doc/sourcebuild.texi (vect_scatter_store): Document.
1962 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1964 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1966 * genopinit.c (main): Add supports_vec_scatter_store and
1967 supports_vec_scatter_store_cached to target_optabs.
1968 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1969 IFN_MASK_SCATTER_STORE.
1970 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1972 * internal-fn.h (internal_store_fn_p): Declare.
1973 (internal_fn_stored_value_index): Likewise.
1974 * internal-fn.c (scatter_store_direct): New macro.
1975 (expand_scatter_store_optab_fn): New function.
1976 (direct_scatter_store_optab_supported_p): New macro.
1977 (internal_store_fn_p): New function.
1978 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1979 IFN_MASK_SCATTER_STORE.
1980 (internal_fn_mask_index): Likewise.
1981 (internal_fn_stored_value_index): New function.
1982 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1984 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1985 * optabs-query.c (supports_vec_scatter_store_p): New function.
1986 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1987 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1988 true for scatter stores.
1989 (vect_gather_scatter_fn_p): Handle scatter stores too.
1990 (vect_check_gather_scatter): Consider using scatter stores if
1991 supports_vec_scatter_store_p.
1992 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1994 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1995 internal_fn_stored_value_index.
1996 (check_load_store_masking): Handle scatter stores too.
1997 (vect_get_store_rhs): Make public.
1998 (vectorizable_call): Use internal_store_fn_p.
1999 (vectorizable_store): Handle scatter store internal functions.
2000 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2001 when deciding whether the end of the group has been reached.
2002 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2003 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2004 (mask_scatter_store<mode>): New insns.
2006 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2007 Alan Hayward <alan.hayward@arm.com>
2008 David Sherwood <david.sherwood@arm.com>
2010 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2011 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2012 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2014 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2015 Use vect_truncate_gather_scatter_offset if we can't treat the
2016 operation as a normal gather load or scatter store.
2017 (get_group_load_store_type): Take the gather_scatter_info
2018 as argument. Try using a gather load or scatter store for
2019 single-element groups.
2020 (get_load_store_type): Update calls to get_group_load_store_type
2021 and vect_use_strided_gather_scatters_p.
2023 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2024 Alan Hayward <alan.hayward@arm.com>
2025 David Sherwood <david.sherwood@arm.com>
2027 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2028 optional tree argument.
2029 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2031 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2032 but continue to use the current value as a fallback.
2033 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2034 to compare the updates.
2035 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2036 (get_load_store_type): Use it when handling a strided access.
2037 (vect_get_strided_load_store_ops): New function.
2038 (vect_get_data_ptr_increment): Likewise.
2039 (vectorizable_load): Handle strided gather loads. Always pass
2040 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2042 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2043 Alan Hayward <alan.hayward@arm.com>
2044 David Sherwood <david.sherwood@arm.com>
2046 * doc/md.texi (gather_load@var{m}): Document.
2047 (mask_gather_load@var{m}): Likewise.
2048 * genopinit.c (main): Add supports_vec_gather_load and
2049 supports_vec_gather_load_cached to target_optabs.
2050 * optabs-tree.c (init_tree_optimization_optabs): Use
2051 ggc_cleared_alloc to allocate target_optabs.
2052 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2053 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2055 * internal-fn.h (internal_load_fn_p): Declare.
2056 (internal_gather_scatter_fn_p): Likewise.
2057 (internal_fn_mask_index): Likewise.
2058 (internal_gather_scatter_fn_supported_p): Likewise.
2059 * internal-fn.c (gather_load_direct): New macro.
2060 (expand_gather_load_optab_fn): New function.
2061 (direct_gather_load_optab_supported_p): New macro.
2062 (direct_internal_fn_optab): New function.
2063 (internal_load_fn_p): Likewise.
2064 (internal_gather_scatter_fn_p): Likewise.
2065 (internal_fn_mask_index): Likewise.
2066 (internal_gather_scatter_fn_supported_p): Likewise.
2067 * optabs-query.c (supports_at_least_one_mode_p): New function.
2068 (supports_vec_gather_load_p): Likewise.
2069 * optabs-query.h (supports_vec_gather_load_p): Declare.
2070 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2071 and memory_type field.
2072 (NUM_PATTERNS): Bump to 15.
2073 * tree-vect-data-refs.c: Include internal-fn.h.
2074 (vect_gather_scatter_fn_p): New function.
2075 (vect_describe_gather_scatter_call): Likewise.
2076 (vect_check_gather_scatter): Try using internal functions for
2077 gather loads. Recognize existing calls to a gather load function.
2078 (vect_analyze_data_refs): Consider using gather loads if
2079 supports_vec_gather_load_p.
2080 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2081 (vect_get_gather_scatter_offset_type): Likewise.
2082 (vect_convert_mask_for_vectype): Likewise.
2083 (vect_add_conversion_to_patterm): Likewise.
2084 (vect_try_gather_scatter_pattern): Likewise.
2085 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2086 (vect_vect_recog_func_ptrs): Add it.
2087 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2088 internal_fn_mask_index and internal_gather_scatter_fn_p.
2089 (check_load_store_masking): Take the gather_scatter_info as an
2090 argument and handle gather loads.
2091 (vect_get_gather_scatter_ops): New function.
2092 (vectorizable_call): Check internal_load_fn_p.
2093 (vectorizable_load): Likewise. Handle gather load internal
2095 (vectorizable_store): Update call to check_load_store_masking.
2096 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2097 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2098 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2099 (aarch64_gather_scale_operand_d): New predicates.
2100 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2101 (mask_gather_load<mode>): New insns.
2103 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2104 Alan Hayward <alan.hayward@arm.com>
2105 David Sherwood <david.sherwood@arm.com>
2107 * optabs.def (fold_left_plus_optab): New optab.
2108 * doc/md.texi (fold_left_plus_@var{m}): Document.
2109 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2110 * internal-fn.c (fold_left_direct): Define.
2111 (expand_fold_left_optab_fn): Likewise.
2112 (direct_fold_left_optab_supported_p): Likewise.
2113 * fold-const-call.c (fold_const_fold_left): New function.
2114 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2115 * tree-parloops.c (valid_reduction_p): New function.
2116 (gather_scalar_reductions): Use it.
2117 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2118 (vect_finish_replace_stmt): Declare.
2119 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2120 (needs_fold_left_reduction_p): New function, split out from...
2121 (vect_is_simple_reduction): ...here. Accept reductions that
2122 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2123 (vect_force_simple_reduction): Also store the reduction type in
2124 the assignment's STMT_VINFO_REDUC_TYPE.
2125 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2126 (merge_with_identity): New function.
2127 (vect_expand_fold_left): Likewise.
2128 (vectorize_fold_left_reduction): Likewise.
2129 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2130 scalar phi in place for it. Check for target support and reject
2131 cases that would reassociate the operation. Defer the transform
2132 phase to vectorize_fold_left_reduction.
2133 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2134 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2135 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2137 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2139 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2140 call to ifc_temp_var.
2142 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2143 Alan Hayward <alan.hayward@arm.com>
2144 David Sherwood <david.sherwood@arm.com>
2146 * target.def (legitimize_address_displacement): Take the original
2147 offset as a poly_int.
2148 * targhooks.h (default_legitimize_address_displacement): Update
2150 * targhooks.c (default_legitimize_address_displacement): Likewise.
2151 * doc/tm.texi: Regenerate.
2152 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2153 as an argument, moving assert of ad->disp == ad->disp_term to...
2154 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2155 Try calling targetm.legitimize_address_displacement before expanding
2156 the address rather than afterwards, and adjust for the new interface.
2157 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2158 Match the new hook interface. Handle SVE addresses.
2159 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2162 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2164 * Makefile.in (OBJS): Add early-remat.o.
2165 * target.def (select_early_remat_modes): New hook.
2166 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2167 * doc/tm.texi: Regenerate.
2168 * targhooks.h (default_select_early_remat_modes): Declare.
2169 * targhooks.c (default_select_early_remat_modes): New function.
2170 * timevar.def (TV_EARLY_REMAT): New timevar.
2171 * passes.def (pass_early_remat): New pass.
2172 * tree-pass.h (make_pass_early_remat): Declare.
2173 * early-remat.c: New file.
2174 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2176 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2178 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2179 Alan Hayward <alan.hayward@arm.com>
2180 David Sherwood <david.sherwood@arm.com>
2182 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2183 vfm1 with a bound_epilog parameter.
2184 (vect_do_peeling): Update calls accordingly, and move the prologue
2185 call earlier in the function. Treat the base bound_epilog as 0 for
2186 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2187 this base when peeling for gaps.
2188 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2189 with fully-masked loops.
2190 (vect_estimate_min_profitable_iters): Handle the single peeled
2191 iteration in that case.
2193 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2194 Alan Hayward <alan.hayward@arm.com>
2195 David Sherwood <david.sherwood@arm.com>
2197 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2198 single-element interleaving even if the size is not a power of 2.
2199 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2200 accesses for single-element interleaving if the group size is
2203 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2204 Alan Hayward <alan.hayward@arm.com>
2205 David Sherwood <david.sherwood@arm.com>
2207 * doc/md.texi (fold_extract_last_@var{m}): Document.
2208 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2209 * optabs.def (fold_extract_last_optab): New optab.
2210 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2211 * internal-fn.c (fold_extract_direct): New macro.
2212 (expand_fold_extract_optab_fn): Likewise.
2213 (direct_fold_extract_optab_supported_p): Likewise.
2214 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2215 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2216 EXTRACT_LAST_REDUCTION.
2217 (get_initial_def_for_reduction): Do not create an initial vector
2218 for EXTRACT_LAST_REDUCTION reductions.
2219 (vectorizable_reduction): Leave the scalar phi in place for
2220 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2221 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2222 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2223 transform phase to vectorizable_condition.
2224 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2226 (vect_finish_stmt_generation): ...here.
2227 (vect_finish_replace_stmt): New function.
2228 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2229 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2231 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2233 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2234 Alan Hayward <alan.hayward@arm.com>
2235 David Sherwood <david.sherwood@arm.com>
2237 * doc/md.texi (extract_last_@var{m}): Document.
2238 * optabs.def (extract_last_optab): New optab.
2239 * internal-fn.def (EXTRACT_LAST): New internal function.
2240 * internal-fn.c (cond_unary_direct): New macro.
2241 (expand_cond_unary_optab_fn): Likewise.
2242 (direct_cond_unary_optab_supported_p): Likewise.
2243 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2244 loops using EXTRACT_LAST.
2245 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2246 (extract_last_<mode>): ...this optab.
2247 (vec_extract<mode><Vel>): Update accordingly.
2249 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2250 Alan Hayward <alan.hayward@arm.com>
2251 David Sherwood <david.sherwood@arm.com>
2253 * target.def (empty_mask_is_expensive): New hook.
2254 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2255 * doc/tm.texi: Regenerate.
2256 * targhooks.h (default_empty_mask_is_expensive): Declare.
2257 * targhooks.c (default_empty_mask_is_expensive): New function.
2258 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2259 if the target says that empty masks are expensive.
2260 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2262 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2264 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2265 Alan Hayward <alan.hayward@arm.com>
2266 David Sherwood <david.sherwood@arm.com>
2268 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2269 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2270 (vect_use_loop_mask_for_alignment_p): New function.
2271 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2272 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2273 niters_skip argument. Make sure that the first niters_skip elements
2274 of the first iteration are inactive.
2275 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2276 Update call to vect_set_loop_masks_directly.
2277 (get_misalign_in_elems): New function, split out from...
2278 (vect_gen_prolog_loop_niters): ...here.
2279 (vect_update_init_of_dr): Take a code argument that specifies whether
2280 the adjustment should be added or subtracted.
2281 (vect_update_init_of_drs): Likewise.
2282 (vect_prepare_for_masked_peels): New function.
2283 (vect_do_peeling): Skip prologue peeling if we're using a mask
2284 instead. Update call to vect_update_inits_of_drs.
2285 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2287 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2288 alignment. Do not include the number of peeled iterations in
2289 the minimum threshold in that case.
2290 (vectorizable_induction): Adjust the start value down by
2291 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2292 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2293 Take the number of skipped iterations into account when calculating
2295 * tree-vect-stmts.c (vect_gen_while_not): New function.
2297 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2298 Alan Hayward <alan.hayward@arm.com>
2299 David Sherwood <david.sherwood@arm.com>
2301 * doc/sourcebuild.texi (vect_fully_masked): Document.
2302 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2304 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2306 (vect_analyze_loop_2): ...here. Don't check the vectorization
2307 factor against the number of loop iterations if the loop is
2310 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2311 Alan Hayward <alan.hayward@arm.com>
2312 David Sherwood <david.sherwood@arm.com>
2314 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2315 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2316 (dump_groups): Update accordingly.
2317 (iv_use::mem_type): New member variable.
2318 (address_p): New function.
2319 (record_use): Add a mem_type argument and initialize the new
2321 (record_group_use): Add a mem_type argument. Use address_p.
2322 Remove obsolete null checks of base_object. Update call to record_use.
2323 (find_interesting_uses_op): Update call to record_group_use.
2324 (find_interesting_uses_cond): Likewise.
2325 (find_interesting_uses_address): Likewise.
2326 (get_mem_type_for_internal_fn): New function.
2327 (find_address_like_use): Likewise.
2328 (find_interesting_uses_stmt): Try find_address_like_use before
2329 calling find_interesting_uses_op.
2330 (addr_offset_valid_p): Use the iv mem_type field as the type
2331 of the addressed memory.
2332 (add_autoinc_candidates): Likewise.
2333 (get_address_cost): Likewise.
2334 (split_small_address_groups_p): Use address_p.
2335 (split_address_groups): Likewise.
2336 (add_iv_candidate_for_use): Likewise.
2337 (autoinc_possible_for_pair): Likewise.
2338 (rewrite_groups): Likewise.
2339 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2340 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2341 (get_alias_ptr_type_for_ptr_address): New function.
2342 (rewrite_use_address): Rewrite address uses in calls that were
2343 identified by find_address_like_use.
2345 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2346 Alan Hayward <alan.hayward@arm.com>
2347 David Sherwood <david.sherwood@arm.com>
2349 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2351 * gimple-expr.h (is_gimple_addressable: Likewise.
2352 * gimple-expr.c (is_gimple_address): Likewise.
2353 * internal-fn.c (expand_call_mem_ref): New function.
2354 (expand_mask_load_optab_fn): Use it.
2355 (expand_mask_store_optab_fn): Likewise.
2357 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2358 Alan Hayward <alan.hayward@arm.com>
2359 David Sherwood <david.sherwood@arm.com>
2361 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2362 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2363 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2364 (cond_umax@var{mode}): Document.
2365 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2366 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2367 (cond_umin_optab, cond_umax_optab): New optabs.
2368 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2369 (COND_IOR, COND_XOR): New internal functions.
2370 * internal-fn.h (get_conditional_internal_fn): Declare.
2371 * internal-fn.c (cond_binary_direct): New macro.
2372 (expand_cond_binary_optab_fn): Likewise.
2373 (direct_cond_binary_optab_supported_p): Likewise.
2374 (get_conditional_internal_fn): New function.
2375 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2376 Cope with reduction statements that are vectorized as calls rather
2378 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2379 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2380 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2381 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2382 (UNSPEC_COND_EOR): New unspecs.
2383 (optab): Add mappings for them.
2384 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2385 (sve_int_op, sve_fp_op): New int attributes.
2387 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2388 Alan Hayward <alan.hayward@arm.com>
2389 David Sherwood <david.sherwood@arm.com>
2391 * optabs.def (while_ult_optab): New optab.
2392 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2393 * internal-fn.def (WHILE_ULT): New internal function.
2394 * internal-fn.h (direct_internal_fn_supported_p): New override
2395 that takes two types as argument.
2396 * internal-fn.c (while_direct): New macro.
2397 (expand_while_optab_fn): New function.
2398 (convert_optab_supported_p): Likewise.
2399 (direct_while_optab_supported_p): New macro.
2400 * wide-int.h (wi::udiv_ceil): New function.
2401 * tree-vectorizer.h (rgroup_masks): New structure.
2402 (vec_loop_masks): New typedef.
2403 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2405 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2406 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2407 (vect_max_vf): New function.
2408 (slpeel_make_loop_iterate_ntimes): Delete.
2409 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2410 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2411 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2412 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2413 internal-fn.h, stor-layout.h and optabs-query.h.
2414 (vect_set_loop_mask): New function.
2415 (add_preheader_seq): Likewise.
2416 (add_header_seq): Likewise.
2417 (interleave_supported_p): Likewise.
2418 (vect_maybe_permute_loop_masks): Likewise.
2419 (vect_set_loop_masks_directly): Likewise.
2420 (vect_set_loop_condition_masked): Likewise.
2421 (vect_set_loop_condition_unmasked): New function, split out from
2422 slpeel_make_loop_iterate_ntimes.
2423 (slpeel_make_loop_iterate_ntimes): Rename to..
2424 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2425 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2426 (vect_do_peeling): Update call accordingly.
2427 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2429 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2430 mask_compare_type, can_fully_mask_p and fully_masked_p.
2431 (release_vec_loop_masks): New function.
2432 (_loop_vec_info): Use it to free the loop masks.
2433 (can_produce_all_loop_masks_p): New function.
2434 (vect_get_max_nscalars_per_iter): Likewise.
2435 (vect_verify_full_masking): Likewise.
2436 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2437 retries, and free the mask rgroups before retrying. Check loop-wide
2438 reasons for disallowing fully-masked loops. Make the final decision
2439 about whether use a fully-masked loop or not.
2440 (vect_estimate_min_profitable_iters): Do not assume that peeling
2441 for the number of iterations will be needed for fully-masked loops.
2442 (vectorizable_reduction): Disable fully-masked loops.
2443 (vectorizable_live_operation): Likewise.
2444 (vect_halve_mask_nunits): New function.
2445 (vect_double_mask_nunits): Likewise.
2446 (vect_record_loop_mask): Likewise.
2447 (vect_get_loop_mask): Likewise.
2448 (vect_transform_loop): Handle the case in which the final loop
2449 iteration might handle a partial vector. Call vect_set_loop_condition
2450 instead of slpeel_make_loop_iterate_ntimes.
2451 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2452 (check_load_store_masking): New function.
2453 (prepare_load_store_mask): Likewise.
2454 (vectorizable_store): Handle fully-masked loops.
2455 (vectorizable_load): Likewise.
2456 (supportable_widening_operation): Use vect_halve_mask_nunits for
2458 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2459 (vect_gen_while): New function.
2460 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2461 (aarch64_uqdec<mode>): New insn.
2463 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2464 Alan Hayward <alan.hayward@arm.com>
2465 David Sherwood <david.sherwood@arm.com>
2467 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2468 (reduc_xor_scal_optab): New optabs.
2469 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2470 (reduc_xor_scal_@var{m}): Document.
2471 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2472 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2474 * fold-const-call.c (fold_const_call): Handle them.
2475 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2476 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2477 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2478 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2479 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2480 (UNSPEC_XORV): New unspecs.
2481 (optab): Add entries for them.
2482 (BITWISEV): New int iterator.
2483 (bit_reduc_op): New int attributes.
2485 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2486 Alan Hayward <alan.hayward@arm.com>
2487 David Sherwood <david.sherwood@arm.com>
2489 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2490 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2491 * optabs.def (vec_shl_insert_optab): New optab.
2492 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2493 (duplicate_and_interleave): Likewise.
2494 * tree-vect-loop.c: Include internal-fn.h.
2495 (neutral_op_for_slp_reduction): New function, split out from
2496 get_initial_defs_for_reduction.
2497 (get_initial_def_for_reduction): Handle option 2 for variable-length
2498 vectors by loading the neutral value into a vector and then shifting
2499 the initial value into element 0.
2500 (get_initial_defs_for_reduction): Replace the code argument with
2501 the neutral value calculated by neutral_op_for_slp_reduction.
2502 Use gimple_build_vector for constant-length vectors.
2503 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2504 but the first group_size elements have a neutral value.
2505 Use duplicate_and_interleave otherwise.
2506 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2507 Update call to get_initial_defs_for_reduction. Handle SLP
2508 reductions for variable-length vectors by creating one vector
2509 result for each scalar result, with the elements associated
2510 with other scalar results stubbed out with the neutral value.
2511 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2512 Require IFN_VEC_SHL_INSERT for double reductions on
2513 variable-length vectors, or SLP reductions that have
2514 a neutral value. Require can_duplicate_and_interleave_p
2515 support for variable-length unchained SLP reductions if there
2516 is no neutral value, such as for MIN/MAX reductions. Also require
2517 the number of vector elements to be a multiple of the number of
2518 SLP statements when doing variable-length unchained SLP reductions.
2519 Update call to vect_create_epilog_for_reduction.
2520 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2521 and remove initial values.
2522 (duplicate_and_interleave): Make public.
2523 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2524 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2526 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2527 Alan Hayward <alan.hayward@arm.com>
2528 David Sherwood <david.sherwood@arm.com>
2530 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2531 (can_duplicate_and_interleave_p): New function.
2532 (vect_get_and_check_slp_defs): Take the vector of statements
2533 rather than just the current one. Remove excess parentheses.
2534 Restriction rejectinon of vect_constant_def and vect_external_def
2535 for variable-length vectors to boolean types, or types for which
2536 can_duplicate_and_interleave_p is false.
2537 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2538 (duplicate_and_interleave): New function.
2539 (vect_get_constant_vectors): Use gimple_build_vector for
2540 constant-length vectors and suitable variable-length constant
2541 vectors. Use duplicate_and_interleave for other variable-length
2542 vectors. Don't defer the update when inserting new statements.
2544 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2545 Alan Hayward <alan.hayward@arm.com>
2546 David Sherwood <david.sherwood@arm.com>
2548 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2549 min_profitable_iters doesn't go negative.
2551 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2552 Alan Hayward <alan.hayward@arm.com>
2553 David Sherwood <david.sherwood@arm.com>
2555 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2556 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2557 * optabs.def (vec_mask_load_lanes_optab): New optab.
2558 (vec_mask_store_lanes_optab): Likewise.
2559 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2560 (MASK_STORE_LANES): Likewise.
2561 * internal-fn.c (mask_load_lanes_direct): New macro.
2562 (mask_store_lanes_direct): Likewise.
2563 (expand_mask_load_optab_fn): Handle masked operations.
2564 (expand_mask_load_lanes_optab_fn): New macro.
2565 (expand_mask_store_optab_fn): Handle masked operations.
2566 (expand_mask_store_lanes_optab_fn): New macro.
2567 (direct_mask_load_lanes_optab_supported_p): Likewise.
2568 (direct_mask_store_lanes_optab_supported_p): Likewise.
2569 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2571 (vect_load_lanes_supported): Likewise.
2572 * tree-vect-data-refs.c (strip_conversion): New function.
2573 (can_group_stmts_p): Likewise.
2574 (vect_analyze_data_ref_accesses): Use it instead of checking
2575 for a pair of assignments.
2576 (vect_store_lanes_supported): Take a masked_p parameter.
2577 (vect_load_lanes_supported): Likewise.
2578 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2579 vect_store_lanes_supported and vect_load_lanes_supported.
2580 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2581 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2582 parameter. Don't allow gaps for masked accesses.
2583 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2584 and vect_load_lanes_supported.
2585 (get_load_store_type): Take a masked_p parameter and update
2586 call to get_group_load_store_type.
2587 (vectorizable_store): Update call to get_load_store_type.
2588 Handle IFN_MASK_STORE_LANES.
2589 (vectorizable_load): Update call to get_load_store_type.
2590 Handle IFN_MASK_LOAD_LANES.
2592 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2593 Alan Hayward <alan.hayward@arm.com>
2594 David Sherwood <david.sherwood@arm.com>
2596 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2598 * config/aarch64/aarch64-protos.h
2599 (aarch64_sve_struct_memory_operand_p): Declare.
2600 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2601 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2602 (VPRED, vpred): Handle SVE structure modes.
2603 * config/aarch64/constraints.md (Utx): New constraint.
2604 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2605 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2606 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2607 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2608 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2609 structure modes. Split into pieces after RA.
2610 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2611 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2613 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2614 SVE structure modes.
2615 (aarch64_classify_address): Likewise.
2616 (sizetochar): Move earlier in file.
2617 (aarch64_print_operand): Handle SVE register lists.
2618 (aarch64_array_mode): New function.
2619 (aarch64_sve_struct_memory_operand_p): Likewise.
2620 (TARGET_ARRAY_MODE): Redefine.
2622 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2623 Alan Hayward <alan.hayward@arm.com>
2624 David Sherwood <david.sherwood@arm.com>
2626 * target.def (array_mode): New target hook.
2627 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2628 * doc/tm.texi: Regenerate.
2629 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2630 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2631 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2633 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2636 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2637 Alan Hayward <alan.hayward@arm.com>
2638 David Sherwood <david.sherwood@arm.com>
2640 * fold-const.c (fold_binary_loc): Check the argument types
2641 rather than the result type when testing for a vector operation.
2643 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2645 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2646 * doc/tm.texi: Regenerate.
2648 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2649 Alan Hayward <alan.hayward@arm.com>
2650 David Sherwood <david.sherwood@arm.com>
2652 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2653 (sve): Document new AArch64 extension.
2654 * doc/md.texi (w): Extend the description of the AArch64
2655 constraint to include SVE vectors.
2656 (Upl, Upa): Document new AArch64 predicate constraints.
2657 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2659 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2660 (msve-vector-bits=): New option.
2661 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2662 SVE when these are disabled.
2663 (sve): New extension.
2664 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2665 modes. Adjust their number of units based on aarch64_sve_vg.
2666 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2667 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2668 aarch64_addr_query_type.
2669 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2670 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2671 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2672 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2673 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2674 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2675 (aarch64_simd_imm_zero_p): Delete.
2676 (aarch64_check_zero_based_sve_index_immediate): Declare.
2677 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2678 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2679 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2680 (aarch64_sve_float_mul_immediate_p): Likewise.
2681 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2683 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2684 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2685 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2686 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2687 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2688 (aarch64_regmode_natural_size): Likewise.
2689 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2690 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2692 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2693 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2694 for VG and the SVE predicate registers.
2695 (V_ALIASES): Add a "z"-prefixed alias.
2696 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2697 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2698 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2699 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2700 (REG_CLASS_NAMES): Add entries for them.
2701 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2702 and the predicate registers.
2703 (aarch64_sve_vg): Declare.
2704 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2705 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2706 (REGMODE_NATURAL_SIZE): Define.
2707 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2709 * config/aarch64/aarch64.c: Include cfgrtl.h.
2710 (simd_immediate_info): Add a constructor for series vectors,
2711 and an associated step field.
2712 (aarch64_sve_vg): New variable.
2713 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2714 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2715 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2716 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2717 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2718 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2719 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2720 (aarch64_get_mask_mode): New functions.
2721 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2722 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2723 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2724 predicate modes and predicate registers. Explicitly restrict
2725 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2726 to store a vector mode if it is recognized by
2727 aarch64_classify_vector_mode.
2728 (aarch64_regmode_natural_size): New function.
2729 (aarch64_hard_regno_caller_save_mode): Return the original mode
2731 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2732 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2733 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2734 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2736 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2737 does not overlap dest if the function is frame-related. Handle
2739 (aarch64_split_add_offset): New function.
2740 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2741 them aarch64_add_offset.
2742 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2743 and update call to aarch64_sub_sp.
2744 (aarch64_add_cfa_expression): New function.
2745 (aarch64_expand_prologue): Pass extra temporary registers to the
2746 functions above. Handle the case in which we need to emit new
2747 DW_CFA_expressions for registers that were originally saved
2748 relative to the stack pointer, but now have to be expressed
2749 relative to the frame pointer.
2750 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2752 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2753 IP0 and IP1 values for SVE frames.
2754 (aarch64_expand_vec_series): New function.
2755 (aarch64_expand_sve_widened_duplicate): Likewise.
2756 (aarch64_expand_sve_const_vector): Likewise.
2757 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2758 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2759 into the register, rather than emitting a SET directly.
2760 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2761 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2762 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2763 (offset_9bit_signed_scaled_p): New functions.
2764 (aarch64_replicate_bitmask_imm): New function.
2765 (aarch64_bitmask_imm): Use it.
2766 (aarch64_cannot_force_const_mem): Reject expressions involving
2767 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2768 (aarch64_classify_index): Handle SVE indices, by requiring
2769 a plain register index with a scale that matches the element size.
2770 (aarch64_classify_address): Handle SVE addresses. Assert that
2771 the mode of the address is VOIDmode or an integer mode.
2772 Update call to aarch64_classify_symbol.
2773 (aarch64_classify_symbolic_expression): Update call to
2774 aarch64_classify_symbol.
2775 (aarch64_const_vec_all_in_range_p): New function.
2776 (aarch64_print_vector_float_operand): Likewise.
2777 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2778 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2779 and the FP immediates 1.0 and 0.5.
2780 (aarch64_print_address_internal): Handle SVE addresses.
2781 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2782 (aarch64_regno_regclass): Handle predicate registers.
2783 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2785 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2786 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2787 (aarch64_convert_sve_vector_bits): New function.
2788 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2789 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2791 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2792 Handle SVE vector and predicate modes. Accept VL-based constants
2793 that need only one temporary register, and VL offsets that require
2794 no temporary registers.
2795 (aarch64_conditional_register_usage): Mark the predicate registers
2796 as fixed if SVE isn't available.
2797 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2798 Return true for SVE vector and predicate modes.
2799 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2800 rather than an unsigned int. Handle SVE modes.
2801 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2803 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2805 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2806 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2807 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2808 (aarch64_sve_float_mul_immediate_p): New functions.
2809 (aarch64_sve_valid_immediate): New function.
2810 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2811 Explicitly reject structure modes. Check for INDEX constants.
2812 Handle PTRUE and PFALSE constants.
2813 (aarch64_check_zero_based_sve_index_immediate): New function.
2814 (aarch64_simd_imm_zero_p): Delete.
2815 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2816 vector modes. Accept constants in the range of CNT[BHWD].
2817 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2818 ask for an Advanced SIMD mode.
2819 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2820 (aarch64_simd_vector_alignment): Handle SVE predicates.
2821 (aarch64_vectorize_preferred_vector_alignment): New function.
2822 (aarch64_simd_vector_alignment_reachable): Use it instead of
2824 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2825 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2827 (MAX_VECT_LEN): Delete.
2828 (expand_vec_perm_d): Add a vec_flags field.
2829 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2830 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2831 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2833 (aarch64_evpc_rev): Rename to...
2834 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2835 (aarch64_evpc_rev_global): New function.
2836 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2837 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2839 (aarch64_evpc_sve_tbl): New function.
2840 (aarch64_expand_vec_perm_const_1): Update after rename of
2841 aarch64_evpc_rev. Handle SVE permutes too, trying
2842 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2843 than aarch64_evpc_tbl.
2844 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2845 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2846 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2847 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2848 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2849 (aarch64_expand_sve_vcond): New functions.
2850 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2851 of aarch64_vector_mode_p.
2852 (aarch64_dwarf_poly_indeterminate_value): New function.
2853 (aarch64_compute_pressure_classes): Likewise.
2854 (aarch64_can_change_mode_class): Likewise.
2855 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2856 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2857 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2858 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2859 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2860 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2861 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2862 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2864 (Dn, Dl, Dr): Accept const as well as const_vector.
2865 (Dz): Likewise. Compare against CONST0_RTX.
2866 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2867 of "vector" where appropriate.
2868 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2869 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2870 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2871 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2872 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2873 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2874 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2875 (v_int_equiv): Extend to SVE modes.
2876 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2878 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2879 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2880 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2881 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2882 (SVE_COND_FP_CMP): New int iterators.
2883 (perm_hilo): Handle the new unpack unspecs.
2884 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2886 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2887 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2888 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2889 (aarch64_equality_operator, aarch64_constant_vector_operand)
2890 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2891 (aarch64_sve_nonimmediate_operand): Likewise.
2892 (aarch64_sve_general_operand): Likewise.
2893 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2894 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2895 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2896 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2897 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2898 (aarch64_sve_float_arith_immediate): Likewise.
2899 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2900 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2901 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2902 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2903 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2904 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2905 (aarch64_sve_float_arith_operand): Likewise.
2906 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2907 (aarch64_sve_float_mul_operand): Likewise.
2908 (aarch64_sve_vec_perm_operand): Likewise.
2909 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2910 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2911 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2912 as well as const_vector.
2913 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2914 in file. Use CONST0_RTX and CONSTM1_RTX.
2915 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2916 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2917 Use aarch64_simd_imm_zero.
2918 * config/aarch64/aarch64-sve.md: New file.
2919 * config/aarch64/aarch64.md: Include it.
2920 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2921 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2922 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2923 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2924 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2925 (sve): New attribute.
2926 (enabled): Disable instructions with the sve attribute unless
2928 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2929 aarch64_expand_mov_immediate.
2930 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2931 CNT[BHSD] immediates.
2932 (movti): Split CONST_POLY_INT moves into two halves.
2933 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2934 Split additions that need a temporary here if the destination
2935 is the stack pointer.
2936 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2937 (*add<mode>3_poly_1): New instruction.
2938 (set_clobber_cc): New expander.
2940 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2942 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2943 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2944 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2945 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2946 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2947 Change innermode from fixed_mode_size to machine_mode.
2948 (simplify_subreg): Update call accordingly. Handle a constant-sized
2949 subreg of a variable-length CONST_VECTOR.
2951 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2952 Alan Hayward <alan.hayward@arm.com>
2953 David Sherwood <david.sherwood@arm.com>
2955 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2956 (add_offset_to_base): New function, split out from...
2957 (create_mem_ref): ...here. When handling a scale other than 1,
2958 check first whether the address is valid without the offset.
2959 Add it into the base if so, leaving the index and scale as-is.
2961 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2964 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2965 fold_for_warn before checking if arg2 is INTEGER_CST.
2967 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2969 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2970 (store_multiple_operation): Delete.
2971 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2972 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2973 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2974 guarded by TARGET_STRING.
2975 (rs6000_output_load_multiple): Delete.
2976 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2977 OPTION_MASK_STRING / TARGET_STRING handling.
2978 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2979 (const rs6000_opt_masks) <"string">: Change mask to 0.
2980 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2981 (MASK_STRING): Delete.
2982 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2984 (load_multiple): Delete.
2991 (store_multiple): Delete.
2998 (movmemsi_8reg): Delete.
2999 (corresponding unnamed define_insn): Delete.
3000 (movmemsi_6reg): Delete.
3001 (corresponding unnamed define_insn): Delete.
3002 (movmemsi_4reg): Delete.
3003 (corresponding unnamed define_insn): Delete.
3004 (movmemsi_2reg): Delete.
3005 (corresponding unnamed define_insn): Delete.
3006 (movmemsi_1reg): Delete.
3007 (corresponding unnamed define_insn): Delete.
3008 * config/rs6000/rs6000.opt (mno-string): New.
3009 (mstring): Replace by deprecation warning stub.
3010 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3012 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3014 * regrename.c (regrename_do_replace): If replacing the same
3015 reg multiple times, try to reuse last created gen_raw_REG.
3018 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3019 main to workaround a bug in GDB.
3021 2018-01-12 Tom de Vries <tom@codesourcery.com>
3024 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3026 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3028 PR rtl-optimization/80481
3029 * ira-color.c (get_cap_member): New function.
3030 (allocnos_conflict_by_live_ranges_p): Use it.
3031 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3032 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3034 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3037 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3038 (*saddl_se_1): Ditto.
3040 (*ssubl_se_1): Ditto.
3042 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3044 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3045 rather than wi::to_widest for DR_INITs.
3046 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3047 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3048 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3050 (vect_analyze_group_access_1): Note that here.
3052 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3054 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3055 polynomial type sizes.
3057 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3059 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3060 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3061 (gimple_add_tmp_var): Likewise.
3063 2018-01-12 Martin Liska <mliska@suse.cz>
3065 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3066 (gimple_alloc_sizes): Likewise.
3067 (dump_gimple_statistics): Use PRIu64 in printf format.
3068 * gimple.h: Change uint64_t to int.
3070 2018-01-12 Martin Liska <mliska@suse.cz>
3072 * tree-core.h: Use uint64_t instead of int.
3073 * tree.c (tree_node_counts): Likewise.
3074 (tree_node_sizes): Likewise.
3075 (dump_tree_statistics): Use PRIu64 in printf format.
3077 2018-01-12 Martin Liska <mliska@suse.cz>
3079 * Makefile.in: As qsort_chk is implemented in vec.c, add
3080 vec.o to linkage of gencfn-macros.
3081 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3082 passing the info to record_node_allocation_statistics.
3083 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3085 * ggc-common.c (struct ggc_usage): Add operator== and use
3086 it in operator< and compare function.
3087 * mem-stats.h (struct mem_usage): Likewise.
3088 * vec.c (struct vec_usage): Remove operator< and compare
3089 function. Can be simply inherited.
3091 2018-01-12 Martin Jambor <mjambor@suse.cz>
3094 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3095 * tree-ssa-math-opts.c: Include domwalk.h.
3096 (convert_mult_to_fma_1): New function.
3097 (fma_transformation_info): New type.
3098 (fma_deferring_state): Likewise.
3099 (cancel_fma_deferring): New function.
3100 (result_of_phi): Likewise.
3101 (last_fma_candidate_feeds_initial_phi): Likewise.
3102 (convert_mult_to_fma): Added deferring logic, split actual
3103 transformation to convert_mult_to_fma_1.
3104 (math_opts_dom_walker): New type.
3105 (math_opts_dom_walker::after_dom_children): New method, body moved
3106 here from pass_optimize_widening_mul::execute, added deferring logic
3108 (pass_optimize_widening_mul::execute): Moved most of code to
3109 math_opts_dom_walker::after_dom_children.
3110 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3111 * config/i386/i386.c (ix86_option_override_internal): Added
3112 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3114 2018-01-12 Richard Biener <rguenther@suse.de>
3117 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3118 inline instance vars.
3120 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3123 * config/rx/rx.c (rx_is_restricted_memory_address):
3126 2018-01-12 Richard Biener <rguenther@suse.de>
3128 PR tree-optimization/80846
3129 * target.def (split_reduction): New target hook.
3130 * targhooks.c (default_split_reduction): New function.
3131 * targhooks.h (default_split_reduction): Declare.
3132 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3133 target requests first reduce vectors by combining low and high
3135 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3136 (get_vectype_for_scalar_type_and_size): Export.
3137 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3138 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3139 * doc/tm.texi: Regenerate.
3140 * config/i386/i386.c (ix86_split_reduction): Implement
3141 TARGET_VECTORIZE_SPLIT_REDUCTION.
3143 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3146 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3147 in PIC mode except for TARGET_VXWORKS_RTP.
3148 * config/sparc/sparc.c: Include cfgrtl.h.
3149 (TARGET_INIT_PIC_REG): Define.
3150 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3151 (sparc_pic_register_p): New predicate.
3152 (sparc_legitimate_address_p): Use it.
3153 (sparc_legitimize_pic_address): Likewise.
3154 (sparc_delegitimize_address): Likewise.
3155 (sparc_mode_dependent_address_p): Likewise.
3156 (gen_load_pcrel_sym): Remove 4th parameter.
3157 (load_got_register): Adjust call to above. Remove obsolete stuff.
3158 (sparc_expand_prologue): Do not call load_got_register here.
3159 (sparc_flat_expand_prologue): Likewise.
3160 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3161 (sparc_use_pseudo_pic_reg): New function.
3162 (sparc_init_pic_reg): Likewise.
3163 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3164 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3166 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3168 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3169 Add item for branch_cost.
3171 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3173 PR rtl-optimization/83565
3174 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3175 not extend the result to a larger mode for rotate operations.
3176 (num_sign_bit_copies1): Likewise.
3178 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3181 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3183 Use values-Xc.o for -pedantic.
3184 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3186 2018-01-12 Martin Liska <mliska@suse.cz>
3189 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3191 (possible_polymorphic_call_targets): Use it.
3192 (ipa_devirt): Likewise.
3194 2018-01-12 Martin Liska <mliska@suse.cz>
3196 * profile-count.h (enum profile_quality): Use 0 as invalid
3197 enum value of profile_quality.
3199 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3201 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3202 -mext-string options.
3204 2018-01-12 Richard Biener <rguenther@suse.de>
3206 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3207 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3208 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3210 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3212 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3214 * configure.ac (--with-long-double-format): Add support for the
3215 configuration option to change the default long double format on
3217 * config.gcc (powerpc*-linux*-*): Likewise.
3218 * configure: Regenerate.
3219 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3220 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3221 used without modification.
3223 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3225 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3226 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3227 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3228 MISC_BUILTIN_SPEC_BARRIER.
3229 (rs6000_init_builtins): Likewise.
3230 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3232 (speculation_barrier): New define_insn.
3233 * doc/extend.texi: Document __builtin_speculation_barrier.
3235 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3238 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3239 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3240 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3242 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3243 integral modes instead of "ss" and "sd".
3244 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3245 vectors with 32-bit and 64-bit elements.
3246 (vecdupssescalarmodesuffix): New mode attribute.
3247 (vec_dup<mode>): Use it.
3249 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3252 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3253 frame if argument is passed on stack.
3255 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3258 * ree.c (combine_reaching_defs): Optimize also
3259 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3260 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3262 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3265 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3267 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3270 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3271 after they are computed.
3273 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3275 PR tree-optimization/83695
3276 * gimple-loop-linterchange.cc
3277 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3278 reset cached scev information after interchange.
3279 (pass_linterchange::execute): Remove call to scev_reset_htab.
3281 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3283 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3284 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3285 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3286 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3287 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3288 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3289 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3290 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3291 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3292 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3293 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3294 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3295 (V_lane_reg): Likewise.
3296 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3298 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3299 (vfmal_lane_low<mode>_intrinsic,
3300 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3301 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3302 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3303 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3304 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3305 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3307 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3309 * config/arm/arm-cpus.in (fp16fml): New feature.
3310 (ALL_SIMD): Add fp16fml.
3311 (armv8.2-a): Add fp16fml as an option.
3312 (armv8.3-a): Likewise.
3313 (armv8.4-a): Add fp16fml as part of fp16.
3314 * config/arm/arm.h (TARGET_FP16FML): Define.
3315 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3317 * config/arm/arm-modes.def (V2HF): Define.
3318 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3319 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3320 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3321 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3322 vfmsl_low, vfmsl_high): New set of builtins.
3323 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3324 (vfml_op): New code attribute.
3325 (VFMLHALVES): New int iterator.
3326 (VFML, VFMLSEL): New mode attributes.
3327 (V_reg): Define mapping for V2HF.
3328 (V_hi, V_lo): New mode attributes.
3329 (VF_constraint): Likewise.
3330 (vfml_half, vfml_half_selector): New int attributes.
3331 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3333 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3334 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3336 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3337 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3338 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3339 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3341 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3342 Document new effective target and option set.
3344 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3346 * config/arm/arm-cpus.in (armv8_4): New feature.
3347 (ARMv8_4a): New fgroup.
3348 (armv8.4-a): New arch.
3349 * config/arm/arm-tables.opt: Regenerate.
3350 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3351 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3352 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3353 Add matching rules for -march=armv8.4-a and extensions.
3354 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3356 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3359 * config/rx/rx.md (BW): New mode attribute.
3360 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3362 2018-01-11 Richard Biener <rguenther@suse.de>
3364 PR tree-optimization/83435
3365 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3366 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3367 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3369 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3370 Alan Hayward <alan.hayward@arm.com>
3371 David Sherwood <david.sherwood@arm.com>
3373 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3375 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3376 (aarch64_print_address_internal): Use it to check for a zero offset.
3378 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3379 Alan Hayward <alan.hayward@arm.com>
3380 David Sherwood <david.sherwood@arm.com>
3382 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3383 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3384 Return a poly_int64 rather than a HOST_WIDE_INT.
3385 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3386 rather than a HOST_WIDE_INT.
3387 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3388 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3389 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3390 final_offset from HOST_WIDE_INT to poly_int64.
3391 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3392 to_constant when getting the number of units in an Advanced SIMD
3394 (aarch64_builtin_vectorized_function): Check for a constant number
3396 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3398 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3399 attribute instead of GET_MODE_NUNITS.
3400 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3401 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3402 GET_MODE_SIZE for fixed-size registers.
3403 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3404 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3405 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3406 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3407 (aarch64_print_operand, aarch64_print_address_internal)
3408 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3409 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3410 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3411 Handle polynomial GET_MODE_SIZE.
3412 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3413 wider than SImode without modification.
3414 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3415 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3416 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3417 passing and returning SVE modes.
3418 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3419 rather than GEN_INT.
3420 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3421 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3422 (aarch64_allocate_and_probe_stack_space): Likewise.
3423 (aarch64_layout_frame): Cope with polynomial offsets.
3424 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3425 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3427 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3428 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3429 poly_int64 rather than a HOST_WIDE_INT.
3430 (aarch64_get_separate_components, aarch64_process_components)
3431 (aarch64_expand_prologue, aarch64_expand_epilogue)
3432 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3433 (aarch64_anchor_offset): New function, split out from...
3434 (aarch64_legitimize_address): ...here.
3435 (aarch64_builtin_vectorization_cost): Handle polynomial
3436 TYPE_VECTOR_SUBPARTS.
3437 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3439 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3440 number of elements from the PARALLEL rather than the mode.
3441 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3442 rather than GET_MODE_BITSIZE.
3443 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3444 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3445 (aarch64_expand_vec_perm_const_1): Handle polynomial
3446 d->perm.length () and d->perm elements.
3447 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3448 Apply to_constant to d->perm elements.
3449 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3450 polynomial CONST_VECTOR_NUNITS.
3451 (aarch64_move_pointer): Take amount as a poly_int64 rather
3453 (aarch64_progress_pointer): Avoid temporary variable.
3454 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3455 the mode attribute instead of GET_MODE.
3457 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3458 Alan Hayward <alan.hayward@arm.com>
3459 David Sherwood <david.sherwood@arm.com>
3461 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3462 x exists before using it.
3463 (aarch64_add_constant_internal): Rename to...
3464 (aarch64_add_offset_1): ...this. Replace regnum with separate
3465 src and dest rtxes. Handle the case in which they're different,
3466 including when the offset is zero. Replace scratchreg with an rtx.
3467 Use 2 additions if there is no spare register into which we can
3468 move a 16-bit constant.
3469 (aarch64_add_constant): Delete.
3470 (aarch64_add_offset): Replace reg with separate src and dest
3471 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3472 Use aarch64_add_offset_1.
3473 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3474 an rtx rather than an int. Take the delta as a poly_int64
3475 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3476 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3477 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3478 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3479 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3481 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3482 aarch64_add_constant.
3484 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3486 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3487 Use scalar_float_mode.
3489 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3491 * config/aarch64/aarch64-simd.md
3492 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3493 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3494 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3495 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3496 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3497 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3498 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3499 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3500 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3501 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3503 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3506 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3507 targ_options->x_arm_arch_string is non NULL.
3509 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3511 * config/aarch64/aarch64.h
3512 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3514 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3517 * expmed.c (emit_store_flag_force): Swap if const op0
3518 and change VOIDmode to mode of op0.
3520 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3522 PR rtl-optimization/83761
3523 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3524 than bytes to mode_for_size.
3526 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3529 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3530 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3533 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3536 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3537 when in layout mode.
3538 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3539 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3542 2018-01-10 Michael Collison <michael.collison@arm.com>
3544 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3545 * config/aarch64/aarch64-option-extension.def: Add
3546 AARCH64_OPT_EXTENSION of 'fp16fml'.
3547 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3548 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3549 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3550 * config/aarch64/constraints.md (Ui7): New constraint.
3551 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3552 (VFMLA_SEL_W): Ditto.
3555 (VFMLA16_LOW): New int iterator.
3556 (VFMLA16_HIGH): Ditto.
3557 (UNSPEC_FMLAL): New unspec.
3558 (UNSPEC_FMLSL): Ditto.
3559 (UNSPEC_FMLAL2): Ditto.
3560 (UNSPEC_FMLSL2): Ditto.
3561 (f16mac): New code attribute.
3562 * config/aarch64/aarch64-simd-builtins.def
3563 (aarch64_fmlal_lowv2sf): Ditto.
3564 (aarch64_fmlsl_lowv2sf): Ditto.
3565 (aarch64_fmlalq_lowv4sf): Ditto.
3566 (aarch64_fmlslq_lowv4sf): Ditto.
3567 (aarch64_fmlal_highv2sf): Ditto.
3568 (aarch64_fmlsl_highv2sf): Ditto.
3569 (aarch64_fmlalq_highv4sf): Ditto.
3570 (aarch64_fmlslq_highv4sf): Ditto.
3571 (aarch64_fmlal_lane_lowv2sf): Ditto.
3572 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3573 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3574 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3575 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3576 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3577 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3578 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3579 (aarch64_fmlal_lane_highv2sf): Ditto.
3580 (aarch64_fmlsl_lane_highv2sf): Ditto.
3581 (aarch64_fmlal_laneq_highv2sf): Ditto.
3582 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3583 (aarch64_fmlalq_lane_highv4sf): Ditto.
3584 (aarch64_fmlsl_lane_highv4sf): Ditto.
3585 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3586 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3587 * config/aarch64/aarch64-simd.md:
3588 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3589 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3590 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3591 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3592 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3593 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3594 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3595 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3596 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3597 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3598 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3599 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3600 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3601 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3602 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3603 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3604 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3605 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3606 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3607 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3608 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3609 (vfmlsl_low_u32): Ditto.
3610 (vfmlalq_low_u32): Ditto.
3611 (vfmlslq_low_u32): Ditto.
3612 (vfmlal_high_u32): Ditto.
3613 (vfmlsl_high_u32): Ditto.
3614 (vfmlalq_high_u32): Ditto.
3615 (vfmlslq_high_u32): Ditto.
3616 (vfmlal_lane_low_u32): Ditto.
3617 (vfmlsl_lane_low_u32): Ditto.
3618 (vfmlal_laneq_low_u32): Ditto.
3619 (vfmlsl_laneq_low_u32): Ditto.
3620 (vfmlalq_lane_low_u32): Ditto.
3621 (vfmlslq_lane_low_u32): Ditto.
3622 (vfmlalq_laneq_low_u32): Ditto.
3623 (vfmlslq_laneq_low_u32): Ditto.
3624 (vfmlal_lane_high_u32): Ditto.
3625 (vfmlsl_lane_high_u32): Ditto.
3626 (vfmlal_laneq_high_u32): Ditto.
3627 (vfmlsl_laneq_high_u32): Ditto.
3628 (vfmlalq_lane_high_u32): Ditto.
3629 (vfmlslq_lane_high_u32): Ditto.
3630 (vfmlalq_laneq_high_u32): Ditto.
3631 (vfmlslq_laneq_high_u32): Ditto.
3632 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3633 (AARCH64_FL_FOR_ARCH8_4): New.
3634 (AARCH64_ISA_F16FML): New ISA flag.
3635 (TARGET_F16FML): New feature flag for fp16fml.
3636 (doc/invoke.texi): Document new fp16fml option.
3638 2018-01-10 Michael Collison <michael.collison@arm.com>
3640 * config/aarch64/aarch64-builtins.c:
3641 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3642 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3643 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3644 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3645 (AARCH64_ISA_SHA3): New ISA flag.
3646 (TARGET_SHA3): New feature flag for sha3.
3647 * config/aarch64/iterators.md (sha512_op): New int attribute.
3648 (CRYPTO_SHA512): New int iterator.
3649 (UNSPEC_SHA512H): New unspec.
3650 (UNSPEC_SHA512H2): Ditto.
3651 (UNSPEC_SHA512SU0): Ditto.
3652 (UNSPEC_SHA512SU1): Ditto.
3653 * config/aarch64/aarch64-simd-builtins.def
3654 (aarch64_crypto_sha512hqv2di): New builtin.
3655 (aarch64_crypto_sha512h2qv2di): Ditto.
3656 (aarch64_crypto_sha512su0qv2di): Ditto.
3657 (aarch64_crypto_sha512su1qv2di): Ditto.
3658 (aarch64_eor3qv8hi): Ditto.
3659 (aarch64_rax1qv2di): Ditto.
3660 (aarch64_xarqv2di): Ditto.
3661 (aarch64_bcaxqv8hi): Ditto.
3662 * config/aarch64/aarch64-simd.md:
3663 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3664 (aarch64_crypto_sha512su0qv2di): Ditto.
3665 (aarch64_crypto_sha512su1qv2di): Ditto.
3666 (aarch64_eor3qv8hi): Ditto.
3667 (aarch64_rax1qv2di): Ditto.
3668 (aarch64_xarqv2di): Ditto.
3669 (aarch64_bcaxqv8hi): Ditto.
3670 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3671 (vsha512h2q_u64): Ditto.
3672 (vsha512su0q_u64): Ditto.
3673 (vsha512su1q_u64): Ditto.
3674 (veor3q_u16): Ditto.
3675 (vrax1q_u64): Ditto.
3677 (vbcaxq_u16): Ditto.
3678 * config/arm/types.md (crypto_sha512): New type attribute.
3679 (crypto_sha3): Ditto.
3680 (doc/invoke.texi): Document new sha3 option.
3682 2018-01-10 Michael Collison <michael.collison@arm.com>
3684 * config/aarch64/aarch64-builtins.c:
3685 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3686 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3687 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3688 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3689 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3690 (AARCH64_ISA_SM4): New ISA flag.
3691 (TARGET_SM4): New feature flag for sm4.
3692 * config/aarch64/aarch64-simd-builtins.def
3693 (aarch64_sm3ss1qv4si): Ditto.
3694 (aarch64_sm3tt1aq4si): Ditto.
3695 (aarch64_sm3tt1bq4si): Ditto.
3696 (aarch64_sm3tt2aq4si): Ditto.
3697 (aarch64_sm3tt2bq4si): Ditto.
3698 (aarch64_sm3partw1qv4si): Ditto.
3699 (aarch64_sm3partw2qv4si): Ditto.
3700 (aarch64_sm4eqv4si): Ditto.
3701 (aarch64_sm4ekeyqv4si): Ditto.
3702 * config/aarch64/aarch64-simd.md:
3703 (aarch64_sm3ss1qv4si): Ditto.
3704 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3705 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3706 (aarch64_sm4eqv4si): Ditto.
3707 (aarch64_sm4ekeyqv4si): Ditto.
3708 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3709 (sm3part_op): Ditto.
3710 (CRYPTO_SM3TT): Ditto.
3711 (CRYPTO_SM3PART): Ditto.
3712 (UNSPEC_SM3SS1): New unspec.
3713 (UNSPEC_SM3TT1A): Ditto.
3714 (UNSPEC_SM3TT1B): Ditto.
3715 (UNSPEC_SM3TT2A): Ditto.
3716 (UNSPEC_SM3TT2B): Ditto.
3717 (UNSPEC_SM3PARTW1): Ditto.
3718 (UNSPEC_SM3PARTW2): Ditto.
3719 (UNSPEC_SM4E): Ditto.
3720 (UNSPEC_SM4EKEY): Ditto.
3721 * config/aarch64/constraints.md (Ui2): New constraint.
3722 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3723 * config/arm/types.md (crypto_sm3): New type attribute.
3724 (crypto_sm4): Ditto.
3725 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3726 (vsm3tt1aq_u32): Ditto.
3727 (vsm3tt1bq_u32): Ditto.
3728 (vsm3tt2aq_u32): Ditto.
3729 (vsm3tt2bq_u32): Ditto.
3730 (vsm3partw1q_u32): Ditto.
3731 (vsm3partw2q_u32): Ditto.
3732 (vsm4eq_u32): Ditto.
3733 (vsm4ekeyq_u32): Ditto.
3734 (doc/invoke.texi): Document new sm4 option.
3736 2018-01-10 Michael Collison <michael.collison@arm.com>
3738 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3739 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3740 (AARCH64_FL_FOR_ARCH8_4): New.
3741 (AARCH64_FL_V8_4): New flag.
3742 (doc/invoke.texi): Document new armv8.4-a option.
3744 2018-01-10 Michael Collison <michael.collison@arm.com>
3746 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3747 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3748 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3749 * config/aarch64/aarch64-option-extension.def: Add
3750 AARCH64_OPT_EXTENSION of 'sha2'.
3751 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3752 (crypto): Disable sha2 and aes if crypto disabled.
3753 (crypto): Enable aes and sha2 if enabled.
3754 (simd): Disable sha2 and aes if simd disabled.
3755 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3757 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3758 (TARGET_SHA2): New feature flag for sha2.
3759 (TARGET_AES): New feature flag for aes.
3760 * config/aarch64/aarch64-simd.md:
3761 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3762 conditional on TARGET_AES.
3763 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3764 (aarch64_crypto_sha1hsi): Make pattern conditional
3766 (aarch64_crypto_sha1hv4si): Ditto.
3767 (aarch64_be_crypto_sha1hv4si): Ditto.
3768 (aarch64_crypto_sha1su1v4si): Ditto.
3769 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3770 (aarch64_crypto_sha1su0v4si): Ditto.
3771 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3772 (aarch64_crypto_sha256su0v4si): Ditto.
3773 (aarch64_crypto_sha256su1v4si): Ditto.
3774 (doc/invoke.texi): Document new aes and sha2 options.
3776 2018-01-10 Martin Sebor <msebor@redhat.com>
3778 PR tree-optimization/83781
3779 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3782 2018-01-11 Martin Sebor <msebor@gmail.com>
3783 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3785 PR tree-optimization/83501
3786 PR tree-optimization/81703
3788 * tree-ssa-strlen.c (get_string_cst): Rename...
3789 (get_string_len): ...to this. Handle global constants.
3790 (handle_char_store): Adjust.
3792 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3793 Jim Wilson <jimw@sifive.com>
3795 * config/riscv/riscv-protos.h (riscv_output_return): New.
3796 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3797 (riscv_attribute_table, riscv_output_return),
3798 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3799 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3800 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3802 (riscv_expand_prologue): Add early return for naked function.
3803 (riscv_expand_epilogue): Likewise.
3804 (riscv_function_ok_for_sibcall): Return false for naked function.
3805 (riscv_set_current_function): New.
3806 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3807 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3808 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3809 * doc/extend.texi (RISC-V Function Attributes): New.
3811 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3813 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3814 check for 128-bit long double before checking TCmode.
3815 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3816 128-bit long doubles before checking TFmode or TCmode.
3817 (FLOAT128_IBM_P): Likewise.
3819 2018-01-10 Martin Sebor <msebor@redhat.com>
3821 PR tree-optimization/83671
3822 * builtins.c (c_strlen): Unconditionally return zero for the empty
3824 Use -Warray-bounds for warnings.
3825 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3826 for non-constant array indices with COMPONENT_REF, arrays of
3827 arrays, and pointers to arrays.
3828 (gimple_fold_builtin_strlen): Determine and set length range for
3829 non-constant character arrays.
3831 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3834 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3837 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3839 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3841 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3844 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3845 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3846 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3847 indexed_or_indirect_operand predicate.
3848 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3849 (*vsx_le_perm_load_v8hi): Likewise.
3850 (*vsx_le_perm_load_v16qi): Likewise.
3851 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3852 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3853 (*vsx_le_perm_store_v8hi): Likewise.
3854 (*vsx_le_perm_store_v16qi): Likewise.
3855 (eight unnamed splitters): Likewise.
3857 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3859 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3860 * config/rs6000/emmintrin.h: Likewise.
3861 * config/rs6000/mmintrin.h: Likewise.
3862 * config/rs6000/xmmintrin.h: Likewise.
3864 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3867 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3869 * tree.c (tree_nop_conversion): Return true for location wrapper
3871 (maybe_wrap_with_location): New function.
3872 (selftest::check_strip_nops): New function.
3873 (selftest::test_location_wrappers): New function.
3874 (selftest::tree_c_tests): Call it.
3875 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3876 (maybe_wrap_with_location): New decl.
3877 (EXPR_LOCATION_WRAPPER_P): New macro.
3878 (location_wrapper_p): New inline function.
3879 (tree_strip_any_location_wrapper): New inline function.
3881 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3884 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3885 stack_realign_offset for the largest alignment of stack slot
3887 (ix86_find_max_used_stack_alignment): New function.
3888 (ix86_finalize_stack_frame_flags): Use it. Set
3889 max_used_stack_alignment if we don't realign stack.
3890 * config/i386/i386.h (machine_function): Add
3891 max_used_stack_alignment.
3893 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3895 * config/arm/arm.opt (-mbranch-cost): New option.
3896 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3899 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3902 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3903 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3905 2018-01-10 Richard Biener <rguenther@suse.de>
3908 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3909 early out so it also covers the case where we have a non-NULL
3912 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3914 PR tree-optimization/83753
3915 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3916 for non-strided grouped accesses if the number of elements is 1.
3918 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3921 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3922 * i386.h (TARGET_USE_GATHER): Define.
3923 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3925 2018-01-10 Martin Liska <mliska@suse.cz>
3928 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3929 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3931 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3932 CLEANUP_NO_PARTITIONING is not set.
3934 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3936 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3937 for vectors, as a partial revert of r254296.
3938 * rtl.h (const_vec_p): Delete.
3939 (const_vec_duplicate_p): Don't test for vector CONSTs.
3940 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3941 * expmed.c (make_tree): Likewise.
3944 * common.md (E, F): Use CONSTANT_P instead of checking for
3946 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3947 checking for CONST_VECTOR.
3949 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3952 * predict.c (force_edge_cold): Handle in more sane way edges
3955 2018-01-09 Carl Love <cel@us.ibm.com>
3957 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3959 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3960 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3961 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3962 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3963 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3964 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3965 * config/rs6000/rs6000-protos.h: Add extern defition for
3966 rs6000_generate_float2_double_code.
3967 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3969 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3970 (float2_v2df): Add define_expand.
3972 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3975 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3976 op_mode in the force_to_mode call.
3978 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3980 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3981 instead of checking each element individually.
3982 (aarch64_evpc_uzp): Likewise.
3983 (aarch64_evpc_zip): Likewise.
3984 (aarch64_evpc_ext): Likewise.
3985 (aarch64_evpc_rev): Likewise.
3986 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3987 instead of checking each element individually. Return true without
3989 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3990 whether all selected elements come from the same input, instead of
3991 checking each element individually. Remove calls to gen_rtx_REG,
3992 start_sequence and end_sequence and instead assert that no rtl is
3995 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3997 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3998 order of HIGH and CONST checks.
4000 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4002 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4003 if the destination isn't an SSA_NAME.
4005 2018-01-09 Richard Biener <rguenther@suse.de>
4007 PR tree-optimization/83668
4008 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4010 (canonicalize_loop_form): ... here, renamed from ...
4011 (canonicalize_loop_closed_ssa_form): ... this and amended to
4012 swap successor edges for loop exit blocks to make us use
4013 the RPO order we need for initial schedule generation.
4015 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4017 PR tree-optimization/64811
4018 * match.pd: When optimizing comparisons with Inf, avoid
4019 introducing or losing exceptions from comparisons with NaN.
4021 2018-01-09 Martin Liska <mliska@suse.cz>
4024 * asan.c (shadow_mem_size): Add gcc_assert.
4026 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4028 Don't save registers in main().
4031 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4032 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4033 * config/avr/avr.c (avr_set_current_function): Don't error if
4034 naked, OS_task or OS_main are specified at the same time.
4035 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4037 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4039 * common/config/avr/avr-common.c (avr_option_optimization_table):
4040 Switch on -mmain-is-OS_task for optimizing compilations.
4042 2018-01-09 Richard Biener <rguenther@suse.de>
4044 PR tree-optimization/83572
4045 * graphite.c: Include cfganal.h.
4046 (graphite_transform_loops): Connect infinite loops to exit
4047 and remove fake edges at the end.
4049 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4051 * ipa-inline.c (edge_badness): Revert accidental checkin.
4053 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4056 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4057 symbols; not inline clones.
4059 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4062 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4063 hard registers. Formatting fixes.
4065 PR preprocessor/83722
4066 * gcc.c (try_generate_repro): Pass
4067 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4068 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4071 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4072 Kito Cheng <kito.cheng@gmail.com>
4074 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4075 (riscv_leaf_function_p): Delete.
4076 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4078 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4080 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4082 (do_ifelse): New function.
4083 (do_isel): New function.
4084 (do_sub3): New function.
4085 (do_add3): New function.
4086 (do_load_mask_compare): New function.
4087 (do_overlap_load_compare): New function.
4088 (expand_compare_loop): New function.
4089 (expand_block_compare): Call expand_compare_loop() when appropriate.
4090 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4092 (-mblock-compare-inline-loop-limit): New option.
4094 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4097 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4098 Reverse order of second and third operands in first alternative.
4099 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4100 of first and second elements in UNSPEC_VPERMR vector.
4101 (altivec_expand_vec_perm_le): Likewise.
4103 2018-01-08 Jeff Law <law@redhat.com>
4105 PR rtl-optimizatin/81308
4106 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4107 (process_switch): If group_case_labels makes a change, then set
4109 (pass_convert_switch::execute): If a switch is converted, then
4110 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4112 PR rtl-optimization/81308
4113 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4116 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4118 PR target/83663 - Revert r255946
4119 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4120 generation for cases where splatting a value is not useful.
4121 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4122 across a vec_duplicate and a paradoxical subreg forming a vector
4123 mode to a vec_concat.
4125 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4127 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4128 -march=armv8.3-a variants.
4129 * config/arm/t-multilib: Likewise.
4130 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4132 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4134 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4136 (cceq_ior_compare_complement): Give it a name so I can use it, and
4137 change boolean_or_operator predicate to boolean_operator so it can
4138 be used to generate a crand.
4139 (eqne): New code iterator.
4140 (bd/bd_neg): New code_attrs.
4141 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4142 a single define_insn.
4143 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4144 decrement (bdnzt/bdnzf/bdzt/bdzf).
4145 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4146 with the new names of the branch decrement patterns, and added the
4147 names of the branch decrement conditional patterns.
4149 2018-01-08 Richard Biener <rguenther@suse.de>
4151 PR tree-optimization/83563
4152 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4155 2018-01-08 Richard Biener <rguenther@suse.de>
4158 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4160 2018-01-08 Richard Biener <rguenther@suse.de>
4162 PR tree-optimization/83685
4163 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4164 references to abnormals.
4166 2018-01-08 Richard Biener <rguenther@suse.de>
4169 * dwarf2out.c (output_indirect_strings): Handle empty
4170 skeleton_debug_str_hash.
4171 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4173 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4175 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4176 (emit_store_direct): Likewise.
4177 (arc_trampoline_adjust_address): Likewise.
4178 (arc_asm_trampoline_template): New function.
4179 (arc_initialize_trampoline): Use asm_trampoline_template.
4180 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4181 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4182 * config/arc/arc.md (flush_icache): Delete pattern.
4184 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4186 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4187 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4190 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4193 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4194 by not USED_FOR_TARGET.
4195 (make_pass_resolve_sw_modes): Likewise.
4197 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4199 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4202 2018-01-08 Richard Biener <rguenther@suse.de>
4205 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4207 2018-01-08 Richard Biener <rguenther@suse.de>
4210 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4212 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4215 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4216 basic blocks with a small number of successors.
4217 (convert_control_dep_chain_into_preds): Improve handling of
4219 (dump_predicates): Split apart into...
4220 (dump_pred_chain): ...here...
4221 (dump_pred_info): ...and here.
4222 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4223 (can_chain_union_be_invalidated_p): Improve check for invalidation
4225 (uninit_uses_cannot_happen): Avoid unnecessary if
4226 convert_control_dep_chain_into_preds yielded nothing.
4228 2018-01-06 Martin Sebor <msebor@redhat.com>
4230 PR tree-optimization/83640
4231 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4232 subtracting negative offset from size.
4233 (builtin_access::overlap): Adjust offset bounds of the access to fall
4234 within the size of the object if possible.
4236 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4238 PR rtl-optimization/83699
4239 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4240 extract_bit_field_as_subreg to cases in which the extracted
4241 value is also a vector.
4243 * lra-constraints.c (process_alt_operands): Test for the equivalence
4244 substitutions when detecting a possible reload cycle.
4246 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4249 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4250 by default if flag_selective_schedling{,2}. Formatting fixes.
4252 PR rtl-optimization/83682
4253 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4254 if it has non-VECTOR_MODE element mode.
4255 (vec_duplicate_p): Likewise.
4258 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4259 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4261 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4264 * config/i386/i386-builtin.def
4265 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4266 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4267 Require also OPTION_MASK_ISA_AVX512F in addition to
4268 OPTION_MASK_ISA_GFNI.
4269 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4270 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4271 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4272 to OPTION_MASK_ISA_GFNI.
4273 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4274 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4275 OPTION_MASK_ISA_AVX512BW.
4276 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4277 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4278 addition to OPTION_MASK_ISA_GFNI.
4279 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4280 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4281 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4282 to OPTION_MASK_ISA_GFNI.
4283 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4284 a requirement for all ISAs rather than any of them with a few
4286 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4288 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4289 bitmasks to be enabled with 3 exceptions, instead of requiring any
4290 enabled ISA with lots of exceptions.
4291 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4292 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4293 Change avx512bw in isa attribute to avx512f.
4294 * config/i386/sgxintrin.h: Add license boilerplate.
4295 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4296 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4297 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4298 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4300 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4301 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4302 temporarily sse2 rather than sse if not enabled already.
4305 * config/i386/sse.md (VI248_VLBW): Rename to ...
4306 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4307 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4308 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4309 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4310 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4311 mode iterator instead of VI248_VLBW.
4313 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4315 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4316 (record_modified): Skip clobbers; add debug output.
4317 (param_change_prob): Use sreal frequencies.
4319 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4321 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4322 punt for user-aligned variables.
4324 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4326 * tree-chrec.c (chrec_contains_symbols): Return true for
4329 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4332 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4333 of (x|y) == x for BICS pattern.
4335 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4337 PR tree-optimization/83605
4338 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4339 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4342 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4344 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4345 * config/epiphany/rtems.h: New file.
4347 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4348 Uros Bizjak <ubizjak@gmail.com>
4351 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4352 QIreg_operand instead of register_operand predicate.
4353 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4354 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4355 comments instead of -fmitigate[-_]rop.
4357 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4360 * cgraphunit.c (symbol_table::compile): Switch to text_section
4361 before calling assembly_start debug hook.
4362 * run-rtl-passes.c (run_rtl_passes): Likewise.
4365 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4367 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4368 range_int_cst_p rather than !symbolic_range_p before calling
4369 extract_range_from_multiplicative_op_1.
4371 2018-01-04 Jeff Law <law@redhat.com>
4373 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4374 redundant test in assertion.
4376 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4378 * doc/rtl.texi: Document machine_mode wrapper classes.
4380 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4382 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4385 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4387 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4388 the VEC_PERM_EXPR fold to fail.
4390 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4393 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4394 to switched_sections.
4396 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4399 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4402 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4405 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4406 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4408 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4411 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4412 is BLKmode and bitpos not zero or mode change is needed.
4414 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4417 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4420 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4423 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4424 instead of MULT rtx. Update all corresponding splitters.
4426 (*ssub<modesuffix>): Ditto.
4428 (*cmp_sadd_di): Update split patterns.
4429 (*cmp_sadd_si): Ditto.
4430 (*cmp_sadd_sidi): Ditto.
4431 (*cmp_ssub_di): Ditto.
4432 (*cmp_ssub_si): Ditto.
4433 (*cmp_ssub_sidi): Ditto.
4434 * config/alpha/predicates.md (const23_operand): New predicate.
4435 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4436 Look for ASHIFT, not MULT inner operand.
4437 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4439 2018-01-04 Martin Liska <mliska@suse.cz>
4441 PR gcov-profile/83669
4442 * gcov.c (output_intermediate_file): Add version to intermediate
4444 * doc/gcov.texi: Document new field 'version' in intermediate
4445 file format. Fix location of '-k' option of gcov command.
4447 2018-01-04 Martin Liska <mliska@suse.cz>
4450 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4452 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4454 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4456 2018-01-03 Martin Sebor <msebor@redhat.com>
4458 PR tree-optimization/83655
4459 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4460 checking calls with invalid arguments.
4462 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4464 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4465 (vectorizable_mask_load_store): Delete.
4466 (vectorizable_call): Return false for masked loads and stores.
4467 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4468 instead of gimple_assign_rhs1.
4469 (vectorizable_load): Handle IFN_MASK_LOAD.
4470 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4474 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4476 (vectorizable_mask_load_store): ...here.
4477 (vectorizable_load): ...and here.
4479 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4481 * tree-vect-stmts.c (vect_build_all_ones_mask)
4482 (vect_build_zero_merge_argument): New functions, split out from...
4483 (vectorizable_load): ...here.
4485 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4487 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4489 (vectorizable_mask_load_store): ...here.
4490 (vectorizable_store): ...and here.
4492 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4494 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4496 (vectorizable_mask_load_store): ...here.
4498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4500 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4501 (vect_model_store_cost): Take a vec_load_store_type instead of a
4503 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4504 (vect_model_store_cost): Take a vec_load_store_type instead of a
4506 (vectorizable_mask_load_store): Update accordingly.
4507 (vectorizable_store): Likewise.
4508 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4510 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4512 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4513 IFN_MASK_LOAD calls here rather than...
4514 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4516 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4517 Alan Hayward <alan.hayward@arm.com>
4518 David Sherwood <david.sherwood@arm.com>
4520 * expmed.c (extract_bit_field_1): For vector extracts,
4521 fall back to extract_bit_field_as_subreg if vec_extract
4524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4525 Alan Hayward <alan.hayward@arm.com>
4526 David Sherwood <david.sherwood@arm.com>
4528 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4529 they are variable or constant sized.
4530 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4531 slots for constant-sized data.
4533 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4534 Alan Hayward <alan.hayward@arm.com>
4535 David Sherwood <david.sherwood@arm.com>
4537 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4538 handling COND_EXPRs with boolean comparisons, try to find a better
4539 basis for the mask type than the boolean itself.
4541 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4543 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4544 is calculated and how it can be overridden.
4545 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4546 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4548 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4551 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4552 Alan Hayward <alan.hayward@arm.com>
4553 David Sherwood <david.sherwood@arm.com>
4555 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4556 Remove the mode argument.
4557 (aarch64_simd_valid_immediate): Remove the mode and inverse
4559 * config/aarch64/iterators.md (bitsize): New iterator.
4560 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4561 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4562 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4563 aarch64_simd_valid_immediate.
4564 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4565 (aarch64_reg_or_bic_imm): Likewise.
4566 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4567 with an insn_type enum and msl with a modifier_type enum.
4568 Replace element_width with a scalar_mode. Change the shift
4569 to unsigned int. Add constructors for scalar_float_mode and
4570 scalar_int_mode elements.
4571 (aarch64_vect_float_const_representable_p): Delete.
4572 (aarch64_can_const_movi_rtx_p)
4573 (aarch64_simd_scalar_immediate_valid_for_move)
4574 (aarch64_simd_make_constant): Update call to
4575 aarch64_simd_valid_immediate.
4576 (aarch64_advsimd_valid_immediate_hs): New function.
4577 (aarch64_advsimd_valid_immediate): Likewise.
4578 (aarch64_simd_valid_immediate): Remove mode and inverse
4579 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4580 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4581 and aarch64_float_const_representable_p on the result.
4582 (aarch64_output_simd_mov_immediate): Remove mode argument.
4583 Update call to aarch64_simd_valid_immediate and use of
4584 simd_immediate_info.
4585 (aarch64_output_scalar_simd_mov_immediate): Update call
4588 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4589 Alan Hayward <alan.hayward@arm.com>
4590 David Sherwood <david.sherwood@arm.com>
4592 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4593 (mode_nunits): Likewise CONST_MODE_NUNITS.
4594 * machmode.def (ADJUST_NUNITS): Document.
4595 * genmodes.c (mode_data::need_nunits_adj): New field.
4596 (blank_mode): Update accordingly.
4597 (adj_nunits): New variable.
4598 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4600 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4601 listed in adj_nunits.
4602 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4603 listed in adj_nunits. Don't emit case statements for such modes.
4604 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4605 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4606 nothing if adj_nunits is nonnull.
4607 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4608 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4609 (emit_mode_fbit): Update use of print_maybe_const_decl.
4610 (emit_move_size): Likewise. Treat the array as non-const
4612 (emit_mode_adjustments): Handle adj_nunits.
4614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4616 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4617 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4618 (VECTOR_MODES): Use it.
4619 (make_vector_modes): Take the prefix as an argument.
4621 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4622 Alan Hayward <alan.hayward@arm.com>
4623 David Sherwood <david.sherwood@arm.com>
4625 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4626 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4627 for MODE_VECTOR_BOOL.
4628 * machmode.def (VECTOR_BOOL_MODE): Document.
4629 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4630 (make_vector_bool_mode): New function.
4631 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4633 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4634 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4636 * stor-layout.c (int_mode_for_mode): Likewise.
4637 * tree.c (build_vector_type_for_mode): Likewise.
4638 * varasm.c (output_constant_pool_2): Likewise.
4639 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4640 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4641 for MODE_VECTOR_BOOL.
4642 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4643 of mode class checks.
4644 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4645 instead of a list of mode class checks.
4646 (expand_vector_scalar_condition): Likewise.
4647 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4650 Alan Hayward <alan.hayward@arm.com>
4651 David Sherwood <david.sherwood@arm.com>
4653 * machmode.h (mode_size): Change from unsigned short to
4655 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4656 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4657 or if measurement_type is not polynomial.
4658 (fixed_size_mode::includes_p): Check for constant-sized modes.
4659 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4660 return a poly_uint16 rather than an unsigned short.
4661 (emit_mode_size): Change the type of mode_size from unsigned short
4662 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4663 (emit_mode_adjustments): Cope with polynomial vector sizes.
4664 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4666 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4668 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4669 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4670 * caller-save.c (setup_save_areas): Likewise.
4671 (replace_reg_with_saved_mem): Likewise.
4672 * calls.c (emit_library_call_value_1): Likewise.
4673 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4674 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4675 (gen_lowpart_for_combine): Likewise.
4676 * convert.c (convert_to_integer_1): Likewise.
4677 * cse.c (equiv_constant, cse_insn): Likewise.
4678 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4679 (cselib_subst_to_values): Likewise.
4680 * dce.c (word_dce_process_block): Likewise.
4681 * df-problems.c (df_word_lr_mark_ref): Likewise.
4682 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4683 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4684 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4685 (rtl_for_decl_location): Likewise.
4686 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4687 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4688 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4689 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4690 (expand_expr_real_1): Likewise.
4691 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4692 (pad_below): Likewise.
4693 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4694 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4695 * ira.c (get_subreg_tracking_sizes): Likewise.
4696 * ira-build.c (ira_create_allocno_objects): Likewise.
4697 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4698 (ira_sort_regnos_for_alter_reg): Likewise.
4699 * ira-costs.c (record_operand_costs): Likewise.
4700 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4701 (resolve_simple_move): Likewise.
4702 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4703 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4704 (lra_constraints): Likewise.
4705 (CONST_POOL_OK_P): Reject variable-sized modes.
4706 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4707 (add_pseudo_to_slot, lra_spill): Likewise.
4708 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4709 * optabs-query.c (get_best_extraction_insn): Likewise.
4710 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4711 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4712 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4713 * recog.c (offsettable_address_addr_space_p): Likewise.
4714 * regcprop.c (maybe_mode_change): Likewise.
4715 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4716 * regrename.c (build_def_use): Likewise.
4717 * regstat.c (dump_reg_info): Likewise.
4718 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4719 (find_reloads, find_reloads_subreg_address): Likewise.
4720 * reload1.c (eliminate_regs_1): Likewise.
4721 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4722 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4723 (simplify_binary_operation_1, simplify_subreg): Likewise.
4724 * targhooks.c (default_function_arg_padding): Likewise.
4725 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4726 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4727 (verify_gimple_assign_ternary): Likewise.
4728 * tree-inline.c (estimate_move_cost): Likewise.
4729 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4730 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4731 (get_address_cost_ainc): Likewise.
4732 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4733 (vect_supportable_dr_alignment): Likewise.
4734 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4735 (vectorizable_reduction): Likewise.
4736 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4737 (vectorizable_operation, vectorizable_load): Likewise.
4738 * tree.c (build_same_sized_truth_vector_type): Likewise.
4739 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4740 * var-tracking.c (emit_note_insn_var_location): Likewise.
4741 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4742 (ADDR_VEC_ALIGN): Likewise.
4744 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4745 Alan Hayward <alan.hayward@arm.com>
4746 David Sherwood <david.sherwood@arm.com>
4748 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4750 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4751 or if measurement_type is polynomial.
4752 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4753 * combine.c (make_extraction): Likewise.
4754 * dse.c (find_shift_sequence): Likewise.
4755 * dwarf2out.c (mem_loc_descriptor): Likewise.
4756 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4757 (extract_bit_field, extract_low_bits): Likewise.
4758 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4759 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4760 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4761 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4762 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4763 * reload.c (find_reloads): Likewise.
4764 * reload1.c (alter_reg): Likewise.
4765 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4766 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4767 * tree-if-conv.c (predicate_mem_writes): Likewise.
4768 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4769 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4770 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4771 * valtrack.c (dead_debug_insert_temp): Likewise.
4772 * varasm.c (mergeable_constant_section): Likewise.
4773 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4775 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4776 Alan Hayward <alan.hayward@arm.com>
4777 David Sherwood <david.sherwood@arm.com>
4779 * expr.c (expand_assignment): Cope with polynomial mode sizes
4780 when assigning to a CONCAT.
4782 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4783 Alan Hayward <alan.hayward@arm.com>
4784 David Sherwood <david.sherwood@arm.com>
4786 * machmode.h (mode_precision): Change from unsigned short to
4788 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4790 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4791 or if measurement_type is not polynomial.
4792 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4793 in which the mode is already known to be a scalar_int_mode.
4794 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4795 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4797 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4798 for GET_MODE_PRECISION.
4799 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4800 for GET_MODE_PRECISION.
4801 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4803 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4804 (expand_field_assignment, make_extraction): Likewise.
4805 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4806 (get_last_value): Likewise.
4807 * convert.c (convert_to_integer_1): Likewise.
4808 * cse.c (cse_insn): Likewise.
4809 * expr.c (expand_expr_real_1): Likewise.
4810 * lra-constraints.c (simplify_operand_subreg): Likewise.
4811 * optabs-query.c (can_atomic_load_p): Likewise.
4812 * optabs.c (expand_atomic_load): Likewise.
4813 (expand_atomic_store): Likewise.
4814 * ree.c (combine_reaching_defs): Likewise.
4815 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4816 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4817 * tree.h (type_has_mode_precision_p): Likewise.
4818 * ubsan.c (instrument_si_overflow): Likewise.
4820 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4821 Alan Hayward <alan.hayward@arm.com>
4822 David Sherwood <david.sherwood@arm.com>
4824 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4825 polynomial numbers of units.
4826 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4827 (valid_vector_subparts_p): New function.
4828 (build_vector_type): Remove temporary shim and take the number
4829 of units as a poly_uint64 rather than an int.
4830 (build_opaque_vector_type): Take the number of units as a
4831 poly_uint64 rather than an int.
4832 * tree.c (build_vector_from_ctor): Handle polynomial
4833 TYPE_VECTOR_SUBPARTS.
4834 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4835 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4836 (build_vector_from_val): If the number of units is variable,
4837 use build_vec_duplicate_cst for constant operands and
4838 VEC_DUPLICATE_EXPR otherwise.
4839 (make_vector_type): Remove temporary is_constant ().
4840 (build_vector_type, build_opaque_vector_type): Take the number of
4841 units as a poly_uint64 rather than an int.
4842 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4844 * cfgexpand.c (expand_debug_expr): Likewise.
4845 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4846 (store_constructor, expand_expr_real_1): Likewise.
4847 (const_scalar_mask_from_tree): Likewise.
4848 * fold-const-call.c (fold_const_reduction): Likewise.
4849 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4850 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4851 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4852 (fold_relational_const): Likewise.
4853 (native_interpret_vector): Likewise. Change the size from an
4854 int to an unsigned int.
4855 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4856 TYPE_VECTOR_SUBPARTS.
4857 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4858 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4859 duplicating a non-constant operand into a variable-length vector.
4860 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4861 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4862 * ipa-icf.c (sem_variable::equals): Likewise.
4863 * match.pd: Likewise.
4864 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4865 * print-tree.c (print_node): Likewise.
4866 * stor-layout.c (layout_type): Likewise.
4867 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4868 * tree-cfg.c (verify_gimple_comparison): Likewise.
4869 (verify_gimple_assign_binary): Likewise.
4870 (verify_gimple_assign_ternary): Likewise.
4871 (verify_gimple_assign_single): Likewise.
4872 * tree-pretty-print.c (dump_generic_node): Likewise.
4873 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4874 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4875 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4876 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4877 (vect_shift_permute_load_chain): Likewise.
4878 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4879 (expand_vector_condition, optimize_vector_constructor): Likewise.
4880 (lower_vec_perm, get_compute_type): Likewise.
4881 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4882 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4883 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4884 (vect_recog_mask_conversion_pattern): Likewise.
4885 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4886 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4887 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4888 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4889 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4890 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4891 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4892 (supportable_widening_operation): Likewise.
4893 (supportable_narrowing_operation): Likewise.
4894 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4896 * varasm.c (output_constant): Likewise.
4898 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4899 Alan Hayward <alan.hayward@arm.com>
4900 David Sherwood <david.sherwood@arm.com>
4902 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4903 so that both the length == 3 and length != 3 cases set up their
4904 own permute vectors. Add comments explaining why we know the
4905 number of elements is constant.
4906 (vect_permute_load_chain): Likewise.
4908 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4909 Alan Hayward <alan.hayward@arm.com>
4910 David Sherwood <david.sherwood@arm.com>
4912 * machmode.h (mode_nunits): Change from unsigned char to
4914 (ONLY_FIXED_SIZE_MODES): New macro.
4915 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4916 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4917 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4919 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4920 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4921 or if measurement_type is not polynomial.
4922 * genmodes.c (ZERO_COEFFS): New macro.
4923 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4925 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4926 Use ZERO_COEFFS when emitting initializers.
4927 * data-streamer.h (bp_pack_poly_value): New function.
4928 (bp_unpack_poly_value): Likewise.
4929 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4930 for GET_MODE_NUNITS.
4931 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4932 for GET_MODE_NUNITS.
4933 * tree.c (make_vector_type): Remove temporary shim and make
4934 the real function take the number of units as a poly_uint64
4936 (build_vector_type_for_mode): Handle polynomial nunits.
4937 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4938 * emit-rtl.c (const_vec_series_p_1): Likewise.
4939 (gen_rtx_CONST_VECTOR): Likewise.
4940 * fold-const.c (test_vec_duplicate_folding): Likewise.
4941 * genrecog.c (validate_pattern): Likewise.
4942 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4943 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4944 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4945 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4946 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4947 * rtlanal.c (subreg_get_info): Likewise.
4948 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4949 (vect_grouped_load_supported): Likewise.
4950 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4951 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4952 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4953 (simplify_const_unary_operation, simplify_binary_operation_1)
4954 (simplify_const_binary_operation, simplify_ternary_operation)
4955 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4956 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4957 instead of CONST_VECTOR_NUNITS.
4958 * varasm.c (output_constant_pool_2): Likewise.
4959 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4960 explicit-encoded elements in the XVEC for variable-length vectors.
4962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4964 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4967 Alan Hayward <alan.hayward@arm.com>
4968 David Sherwood <david.sherwood@arm.com>
4970 * coretypes.h (fixed_size_mode): Declare.
4971 (fixed_size_mode_pod): New typedef.
4972 * builtins.h (target_builtins::x_apply_args_mode)
4973 (target_builtins::x_apply_result_mode): Change type to
4974 fixed_size_mode_pod.
4975 * builtins.c (apply_args_size, apply_result_size, result_vector)
4976 (expand_builtin_apply_args_1, expand_builtin_apply)
4977 (expand_builtin_return): Update accordingly.
4979 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4981 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4982 * cselib.c (cselib_hash_rtx): Likewise.
4983 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4984 CONST_VECTOR encoding.
4986 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4987 Jeff Law <law@redhat.com>
4990 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4991 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4992 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4993 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4996 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4997 explicitly probe *sp in a noreturn function if there were any callee
4998 register saves or frame pointer is needed.
5000 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5003 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5004 BLKmode for ternary, binary or unary expressions.
5007 * var-tracking.c (delete_vta_debug_insn): New inline function.
5008 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5009 insns from get_insns () to NULL instead of each bb separately.
5010 Use delete_vta_debug_insn. No longer static.
5011 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5012 delete_vta_debug_insns callers.
5013 * rtl.h (delete_vta_debug_insns): Declare.
5014 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5015 instead of variable_tracking_main.
5017 2018-01-03 Martin Sebor <msebor@redhat.com>
5019 PR tree-optimization/83603
5020 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5021 arguments past the endof the argument list in functions declared
5022 without a prototype.
5023 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5024 Avoid checking when arguments are null.
5026 2018-01-03 Martin Sebor <msebor@redhat.com>
5029 * doc/extend.texi (attribute const): Fix a typo.
5030 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5031 issuing -Wsuggest-attribute for void functions.
5033 2018-01-03 Martin Sebor <msebor@redhat.com>
5035 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5036 offset_int::from instead of wide_int::to_shwi.
5037 (maybe_diag_overlap): Remove assertion.
5038 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5039 * gimple-ssa-sprintf.c (format_directive): Same.
5040 (parse_directive): Same.
5041 (sprintf_dom_walker::compute_format_length): Same.
5042 (try_substitute_return_value): Same.
5044 2018-01-03 Jeff Law <law@redhat.com>
5047 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5048 non-constant residual for zero at runtime and avoid probing in
5049 that case. Reorganize code for trailing problem to mirror handling
5052 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5054 PR tree-optimization/83501
5055 * tree-ssa-strlen.c (get_string_cst): New.
5056 (handle_char_store): Call get_string_cst.
5058 2018-01-03 Martin Liska <mliska@suse.cz>
5060 PR tree-optimization/83593
5061 * tree-ssa-strlen.c: Include tree-cfg.h.
5062 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5063 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5064 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5066 (strlen_dom_walker::before_dom_children): Call
5067 gimple_purge_dead_eh_edges. Dump tranformation with details
5069 (strlen_dom_walker::before_dom_children): Update call by adding
5070 new argument cleanup_eh.
5071 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5073 2018-01-03 Martin Liska <mliska@suse.cz>
5076 * cif-code.def (VARIADIC_THUNK): New enum value.
5077 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5080 2018-01-03 Jan Beulich <jbeulich@suse.com>
5082 * sse.md (mov<mode>_internal): Tighten condition for when to use
5083 vmovdqu<ssescalarsize> for TI and OI modes.
5085 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5087 Update copyright years.
5089 2018-01-03 Martin Liska <mliska@suse.cz>
5092 * ipa-visibility.c (function_and_variable_visibility): Skip
5093 functions with noipa attribure.
5095 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5097 * gcc.c (process_command): Update copyright notice dates.
5098 * gcov-dump.c (print_version): Ditto.
5099 * gcov.c (print_version): Ditto.
5100 * gcov-tool.c (print_version): Ditto.
5101 * gengtype.c (create_file): Ditto.
5102 * doc/cpp.texi: Bump @copying's copyright year.
5103 * doc/cppinternals.texi: Ditto.
5104 * doc/gcc.texi: Ditto.
5105 * doc/gccint.texi: Ditto.
5106 * doc/gcov.texi: Ditto.
5107 * doc/install.texi: Ditto.
5108 * doc/invoke.texi: Ditto.
5110 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5112 * vector-builder.h (vector_builder::m_full_nelts): Change from
5113 unsigned int to poly_uint64.
5114 (vector_builder::full_nelts): Update prototype accordingly.
5115 (vector_builder::new_vector): Likewise.
5116 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5117 (vector_builder::operator ==): Likewise.
5118 (vector_builder::finalize): Likewise.
5119 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5120 Take the number of elements as a poly_uint64 rather than an
5122 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5123 from unsigned int to poly_uint64.
5124 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5125 (vec_perm_indices::new_vector): Likewise.
5126 (vec_perm_indices::length): Likewise.
5127 (vec_perm_indices::nelts_per_input): Likewise.
5128 (vec_perm_indices::input_nelts): Likewise.
5129 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5130 number of elements per input as a poly_uint64 rather than an
5131 unsigned int. Use the original encoding for variable-length
5132 vectors, rather than clamping each individual element.
5133 For the second and subsequent elements in each pattern,
5134 clamp the step and base before clamping their sum.
5135 (vec_perm_indices::series_p): Handle polynomial element counts.
5136 (vec_perm_indices::all_in_range_p): Likewise.
5137 (vec_perm_indices_to_tree): Likewise.
5138 (vec_perm_indices_to_rtx): Likewise.
5139 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5140 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5141 (tree_vector_builder::new_binary_operation): Handle polynomial
5142 element counts. Return false if we need to know the number
5143 of elements at compile time.
5144 * fold-const.c (fold_vec_perm): Punt if the number of elements
5145 isn't known at compile time.
5147 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5149 * vec-perm-indices.h (vec_perm_builder): Change element type
5150 from HOST_WIDE_INT to poly_int64.
5151 (vec_perm_indices::element_type): Update accordingly.
5152 (vec_perm_indices::clamp): Handle polynomial element_types.
5153 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5154 (vec_perm_indices::all_in_range_p): Likewise.
5155 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5157 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5158 polynomial vec_perm_indices element types.
5159 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5160 * fold-const.c (fold_vec_perm): Likewise.
5161 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5162 * tree-vect-generic.c (lower_vec_perm): Likewise.
5163 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5164 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5165 element type to HOST_WIDE_INT.
5167 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5168 Alan Hayward <alan.hayward@arm.com>
5169 David Sherwood <david.sherwood@arm.com>
5171 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5172 rather than an int. Use plus_constant.
5173 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5174 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5176 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5177 Alan Hayward <alan.hayward@arm.com>
5178 David Sherwood <david.sherwood@arm.com>
5180 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5181 a HOST_WIDE_INT to a poly_int64.
5183 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5184 Alan Hayward <alan.hayward@arm.com>
5185 David Sherwood <david.sherwood@arm.com>
5187 * calls.c (load_register_parameters): Cope with polynomial
5188 mode sizes. Require a constant size for BLKmode parameters
5189 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5190 forces a parameter to be padded at the lsb end in order to
5191 fill a complete number of words, require the parameter size
5192 to be ordered wrt UNITS_PER_WORD.
5194 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5195 Alan Hayward <alan.hayward@arm.com>
5196 David Sherwood <david.sherwood@arm.com>
5198 * reload1.c (spill_stack_slot_width): Change element type
5199 from unsigned int to poly_uint64_pod.
5200 (alter_reg): Treat mode sizes as polynomial.
5202 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5203 Alan Hayward <alan.hayward@arm.com>
5204 David Sherwood <david.sherwood@arm.com>
5206 * reload.c (complex_word_subreg_p): New function.
5207 (reload_inner_reg_of_subreg, push_reload): Use it.
5209 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5210 Alan Hayward <alan.hayward@arm.com>
5211 David Sherwood <david.sherwood@arm.com>
5213 * lra-constraints.c (process_alt_operands): Reject matched
5214 operands whose sizes aren't ordered.
5215 (match_reload): Refer to this check here.
5217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5218 Alan Hayward <alan.hayward@arm.com>
5219 David Sherwood <david.sherwood@arm.com>
5221 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5222 that the mode size is in the set {1, 2, 4, 8, 16}.
5224 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5225 Alan Hayward <alan.hayward@arm.com>
5226 David Sherwood <david.sherwood@arm.com>
5228 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5229 Use plus_constant instead of gen_rtx_PLUS.
5231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5232 Alan Hayward <alan.hayward@arm.com>
5233 David Sherwood <david.sherwood@arm.com>
5235 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5236 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5237 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5238 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5239 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5240 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5241 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5242 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5243 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5244 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5246 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5247 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5248 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5249 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5250 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5251 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5252 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5253 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5254 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5255 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5257 * expr.c (emit_move_resolve_push): Treat the input and result
5258 of PUSH_ROUNDING as a poly_int64.
5259 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5260 (emit_push_insn): Likewise.
5261 * lra-eliminations.c (mark_not_eliminable): Likewise.
5262 * recog.c (push_operand): Likewise.
5263 * reload1.c (elimination_effects): Likewise.
5264 * rtlanal.c (nonzero_bits1): Likewise.
5265 * calls.c (store_one_arg): Likewise. Require the padding to be
5266 known at compile time.
5268 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5269 Alan Hayward <alan.hayward@arm.com>
5270 David Sherwood <david.sherwood@arm.com>
5272 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5273 Use plus_constant instead of gen_rtx_PLUS.
5275 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5276 Alan Hayward <alan.hayward@arm.com>
5277 David Sherwood <david.sherwood@arm.com>
5279 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5283 Alan Hayward <alan.hayward@arm.com>
5284 David Sherwood <david.sherwood@arm.com>
5286 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5287 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5288 via stack temporaries. Treat the mode size as polynomial too.
5290 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5291 Alan Hayward <alan.hayward@arm.com>
5292 David Sherwood <david.sherwood@arm.com>
5294 * expr.c (expand_expr_real_2): When handling conversions involving
5295 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5296 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5297 as a poly_uint64 too.
5299 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5300 Alan Hayward <alan.hayward@arm.com>
5301 David Sherwood <david.sherwood@arm.com>
5303 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5306 Alan Hayward <alan.hayward@arm.com>
5307 David Sherwood <david.sherwood@arm.com>
5309 * combine.c (can_change_dest_mode): Handle polynomial
5310 REGMODE_NATURAL_SIZE.
5311 * expmed.c (store_bit_field_1): Likewise.
5312 * expr.c (store_constructor): Likewise.
5313 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5314 and polynomial REGMODE_NATURAL_SIZE.
5315 (gen_lowpart_common): Likewise.
5316 * reginfo.c (record_subregs_of_mode): Likewise.
5317 * rtlanal.c (read_modify_subreg_p): Likewise.
5319 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5320 Alan Hayward <alan.hayward@arm.com>
5321 David Sherwood <david.sherwood@arm.com>
5323 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5324 numbers of elements.
5326 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5327 Alan Hayward <alan.hayward@arm.com>
5328 David Sherwood <david.sherwood@arm.com>
5330 * match.pd: Cope with polynomial numbers of vector elements.
5332 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5333 Alan Hayward <alan.hayward@arm.com>
5334 David Sherwood <david.sherwood@arm.com>
5336 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5337 in a POINTER_PLUS_EXPR.
5339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5340 Alan Hayward <alan.hayward@arm.com>
5341 David Sherwood <david.sherwood@arm.com>
5343 * omp-simd-clone.c (simd_clone_subparts): New function.
5344 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5345 (ipa_simd_modify_function_body): Likewise.
5347 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5348 Alan Hayward <alan.hayward@arm.com>
5349 David Sherwood <david.sherwood@arm.com>
5351 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5352 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5353 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5354 (expand_vector_condition, vector_element): Likewise.
5355 (subparts_gt): New function.
5356 (get_compute_type): Use subparts_gt.
5357 (count_type_subparts): Delete.
5358 (expand_vector_operations_1): Use subparts_gt instead of
5359 count_type_subparts.
5361 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5362 Alan Hayward <alan.hayward@arm.com>
5363 David Sherwood <david.sherwood@arm.com>
5365 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5366 (vect_compile_time_alias): ...this new function. Do the calculation
5367 on poly_ints rather than trees.
5368 (vect_prune_runtime_alias_test_list): Update call accordingly.
5370 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5371 Alan Hayward <alan.hayward@arm.com>
5372 David Sherwood <david.sherwood@arm.com>
5374 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5376 (vect_schedule_slp_instance): Likewise.
5378 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5379 Alan Hayward <alan.hayward@arm.com>
5380 David Sherwood <david.sherwood@arm.com>
5382 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5383 constant and extern definitions for variable-length vectors.
5384 (vect_get_constant_vectors): Note that the number of units
5385 is known to be constant.
5387 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5388 Alan Hayward <alan.hayward@arm.com>
5389 David Sherwood <david.sherwood@arm.com>
5391 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5392 of units as polynomial. Choose between WIDE and NARROW based
5395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5396 Alan Hayward <alan.hayward@arm.com>
5397 David Sherwood <david.sherwood@arm.com>
5399 * tree-vect-stmts.c (simd_clone_subparts): New function.
5400 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5402 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5403 Alan Hayward <alan.hayward@arm.com>
5404 David Sherwood <david.sherwood@arm.com>
5406 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5407 vectors as polynomial. Use build_index_vector for
5410 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5411 Alan Hayward <alan.hayward@arm.com>
5412 David Sherwood <david.sherwood@arm.com>
5414 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5415 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5416 for variable-length vectors.
5417 (vectorizable_mask_load_store): Treat the number of units as
5418 polynomial, asserting that it is constant if the condition has
5419 already been enforced.
5420 (vectorizable_store, vectorizable_load): Likewise.
5422 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5423 Alan Hayward <alan.hayward@arm.com>
5424 David Sherwood <david.sherwood@arm.com>
5426 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5427 of units as polynomial. Punt if we can't tell at compile time
5428 which vector contains the final result.
5430 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5431 Alan Hayward <alan.hayward@arm.com>
5432 David Sherwood <david.sherwood@arm.com>
5434 * tree-vect-loop.c (vectorizable_induction): Treat the number
5435 of units as polynomial. Punt on SLP inductions. Use an integer
5436 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5437 cast of such a series for variable-length floating-point
5440 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5441 Alan Hayward <alan.hayward@arm.com>
5442 David Sherwood <david.sherwood@arm.com>
5444 * tree.h (build_index_vector): Declare.
5445 * tree.c (build_index_vector): New function.
5446 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5447 of units as polynomial, forcibly converting it to a constant if
5448 vectorizable_reduction has already enforced the condition.
5449 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5450 to create a {1,2,3,...} vector.
5451 (vectorizable_reduction): Treat the number of units as polynomial.
5452 Choose vectype_in based on the largest scalar element size rather
5453 than the smallest number of units. Enforce the restrictions
5456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5457 Alan Hayward <alan.hayward@arm.com>
5458 David Sherwood <david.sherwood@arm.com>
5460 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5461 number of units as polynomial.
5463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5464 Alan Hayward <alan.hayward@arm.com>
5465 David Sherwood <david.sherwood@arm.com>
5467 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5468 * target.def (autovectorize_vector_sizes): Return the vector sizes
5469 by pointer, using vector_sizes rather than a bitmask.
5470 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5471 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5472 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5474 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5475 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5476 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5477 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5478 * omp-general.c (omp_max_vf): Likewise.
5479 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5480 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5481 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5482 * tree-vect-slp.c (vect_slp_bb): Likewise.
5483 * doc/tm.texi: Regenerate.
5484 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5486 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5487 the vector size as a poly_uint64 rather than an unsigned int.
5488 (current_vector_size): Change from an unsigned int to a poly_uint64.
5489 (get_vectype_for_scalar_type): Update accordingly.
5490 * tree.h (build_truth_vector_type): Take the size and number of
5491 units as a poly_uint64 rather than an unsigned int.
5492 (build_vector_type): Add a temporary overload that takes
5493 the number of units as a poly_uint64 rather than an unsigned int.
5494 * tree.c (make_vector_type): Likewise.
5495 (build_truth_vector_type): Take the number of units as a poly_uint64
5496 rather than an unsigned int.
5498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5499 Alan Hayward <alan.hayward@arm.com>
5500 David Sherwood <david.sherwood@arm.com>
5502 * target.def (get_mask_mode): Take the number of units and length
5503 as poly_uint64s rather than unsigned ints.
5504 * targhooks.h (default_get_mask_mode): Update accordingly.
5505 * targhooks.c (default_get_mask_mode): Likewise.
5506 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5507 * doc/tm.texi: Regenerate.
5509 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5510 Alan Hayward <alan.hayward@arm.com>
5511 David Sherwood <david.sherwood@arm.com>
5513 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5514 * omp-general.c (omp_max_vf): Likewise.
5515 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5516 (expand_omp_simd): Handle polynomial safelen.
5517 * omp-low.c (omplow_simd_context): Add a default constructor.
5518 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5519 (lower_rec_simd_input_clauses): Update accordingly.
5520 (lower_rec_input_clauses): Likewise.
5522 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5523 Alan Hayward <alan.hayward@arm.com>
5524 David Sherwood <david.sherwood@arm.com>
5526 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5527 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5528 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5529 (vect_analyze_slp_cost): Likewise.
5530 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5531 (vect_model_load_cost): Likewise.
5533 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5534 Alan Hayward <alan.hayward@arm.com>
5535 David Sherwood <david.sherwood@arm.com>
5537 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5538 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5539 from an unsigned int * to a poly_uint64_pod *.
5540 (calculate_unrolling_factor): New function.
5541 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5544 Alan Hayward <alan.hayward@arm.com>
5545 David Sherwood <david.sherwood@arm.com>
5547 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5548 from an unsigned int to a poly_uint64.
5549 (_loop_vec_info::slp_unrolling_factor): Likewise.
5550 (_loop_vec_info::vectorization_factor): Change from an int
5552 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5553 (vect_get_num_vectors): New function.
5554 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5555 (vect_get_num_copies): Use vect_get_num_vectors.
5556 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5557 to an unsigned int *.
5558 (vect_analyze_data_refs): Change min_vf from an int * to a
5560 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5561 than an unsigned HOST_WIDE_INT.
5562 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5563 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5564 to an unsigned int *.
5565 (vect_analyze_data_ref_dependences): Likewise.
5566 (vect_compute_data_ref_alignment): Handle polynomial vf.
5567 (vect_enhance_data_refs_alignment): Likewise.
5568 (vect_prune_runtime_alias_test_list): Likewise.
5569 (vect_shift_permute_load_chain): Likewise.
5570 (vect_supportable_dr_alignment): Likewise.
5571 (dependence_distance_ge_vf): Take the vectorization factor as a
5572 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5573 (vect_analyze_data_refs): Change min_vf from an int * to a
5575 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5576 vfm1 as a poly_uint64 rather than an int. Make the same change
5577 for the returned bound_scalar.
5578 (vect_gen_vector_loop_niters): Handle polynomial vf.
5579 (vect_do_peeling): Likewise. Update call to
5580 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5581 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5583 * tree-vect-loop.c (vect_determine_vectorization_factor)
5584 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5585 (vect_get_known_peeling_cost): Likewise.
5586 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5587 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5588 (vect_transform_loop): Likewise. Use the lowest possible VF when
5589 updating the upper bounds of the loop.
5590 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5592 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5593 polynomial unroll factors.
5594 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5595 (vect_make_slp_decision): Likewise.
5596 (vect_supported_load_permutation_p): Likewise, and polynomial
5598 (vect_analyze_slp_cost): Handle polynomial vf.
5599 (vect_slp_analyze_node_operations): Likewise.
5600 (vect_slp_analyze_bb_1): Likewise.
5601 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5602 than an unsigned HOST_WIDE_INT.
5603 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5604 (vectorizable_load): Handle polynomial vf.
5605 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5607 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5609 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5610 Alan Hayward <alan.hayward@arm.com>
5611 David Sherwood <david.sherwood@arm.com>
5613 * match.pd: Handle bit operations involving three constants
5614 and try to fold one pair.
5616 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5618 * tree-vect-loop-manip.c: Include gimple-fold.h.
5619 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5620 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5621 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5622 Add a path that uses a step of VF instead of 1, but disable it
5624 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5625 and niters_no_overflow parameters. Update calls to
5626 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5627 Create a new SSA name if the latter choses to use a ste other
5628 than zero, and return it via niters_vector_mult_vf_var.
5629 * tree-vect-loop.c (vect_transform_loop): Update calls to
5630 vect_do_peeling, vect_gen_vector_loop_niters and
5631 slpeel_make_loop_iterate_ntimes.
5632 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5633 (vect_gen_vector_loop_niters): Update declarations after above changes.
5635 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5637 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5638 128-bit round to integer instructions.
5639 (ceil<mode>2): Likewise.
5640 (btrunc<mode>2): Likewise.
5641 (round<mode>2): Likewise.
5643 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5645 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5646 unaligned VSX load/store on P8/P9.
5647 (expand_block_clear): Allow the use of unaligned VSX
5648 load/store on P8/P9.
5650 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5652 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5654 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5655 swap associated with both a load and a store.
5657 2018-01-02 Andrew Waterman <andrew@sifive.com>
5659 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5660 * config/riscv/riscv.md (clear_cache): Use it.
5662 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5664 * web.c: Remove out-of-date comment.
5666 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5668 * expr.c (fixup_args_size_notes): Check that any existing
5669 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5670 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5671 (emit_single_push_insn): ...here.
5673 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5675 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5676 (const_vector_encoded_nelts): New function.
5677 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5678 (const_vector_int_elt, const_vector_elt): Declare.
5679 * emit-rtl.c (const_vector_int_elt_1): New function.
5680 (const_vector_elt): Likewise.
5681 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5682 of CONST_VECTOR_ELT.
5684 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5686 * expr.c: Include rtx-vector-builder.h.
5687 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5688 directly on the tree encoding.
5689 (const_vector_from_tree): Likewise.
5690 * optabs.c: Include rtx-vector-builder.h.
5691 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5692 sequence of "u" values.
5693 * vec-perm-indices.c: Include rtx-vector-builder.h.
5694 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5695 directly on the vec_perm_indices encoding.
5697 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5699 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5700 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5701 * rtx-vector-builder.h: New file.
5702 * rtx-vector-builder.c: Likewise.
5703 * rtl.h (rtx_def::u2): Add a const_vector field.
5704 (CONST_VECTOR_NPATTERNS): New macro.
5705 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5706 (CONST_VECTOR_DUPLICATE_P): Likewise.
5707 (CONST_VECTOR_STEPPED_P): Likewise.
5708 (CONST_VECTOR_ENCODED_ELT): Likewise.
5709 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5710 (unwrap_const_vec_duplicate): Likewise.
5711 (const_vec_series_p): Check for a non-duplicated vector encoding.
5712 Say that the function only returns true for integer vectors.
5713 * emit-rtl.c: Include rtx-vector-builder.h.
5714 (gen_const_vec_duplicate_1): Delete.
5715 (gen_const_vector): Call gen_const_vec_duplicate instead of
5716 gen_const_vec_duplicate_1.
5717 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5718 (gen_const_vec_duplicate): Use rtx_vector_builder.
5719 (gen_const_vec_series): Likewise.
5720 (gen_rtx_CONST_VECTOR): Likewise.
5721 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5722 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5723 Build a new vector rather than modifying a CONST_VECTOR in-place.
5724 (handle_special_swappables): Update call accordingly.
5725 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5726 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5727 Build a new vector rather than modifying a CONST_VECTOR in-place.
5728 (handle_special_swappables): Update call accordingly.
5730 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5732 * simplify-rtx.c (simplify_const_binary_operation): Use
5733 CONST_VECTOR_ELT instead of XVECEXP.
5735 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5737 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5738 the selector elements to be different from the data elements
5739 if the selector is a VECTOR_CST.
5740 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5741 ssizetype for the selector.
5743 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5745 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5746 before testing each element individually.
5747 * tree-vect-generic.c (lower_vec_perm): Likewise.
5749 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5751 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5752 * selftest-run-tests.c (selftest::run_tests): Call it.
5753 * vector-builder.h (vector_builder::operator ==): New function.
5754 (vector_builder::operator !=): Likewise.
5755 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5756 (vec_perm_indices::all_from_input_p): New function.
5757 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5758 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5759 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5760 instead of reading the VECTOR_CST directly. Detect whether both
5761 vector inputs are the same before constructing the vec_perm_indices,
5762 and update the number of inputs argument accordingly. Use the
5763 utility functions added above. Only construct sel2 if we need to.
5765 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5767 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5768 the broadcast of the low byte.
5769 (expand_mult_highpart): Use an explicit encoding for the permutes.
5770 * optabs-query.c (can_mult_highpart_p): Likewise.
5771 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5772 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5773 (vectorizable_bswap): Likewise.
5774 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5775 explicit encoding for the power-of-2 permutes.
5776 (vect_permute_store_chain): Likewise.
5777 (vect_grouped_load_supported): Likewise.
5778 (vect_permute_load_chain): Likewise.
5780 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5782 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5783 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5784 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5785 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5786 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5787 (vect_gen_perm_mask_any): Likewise.
5789 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5791 * int-vector-builder.h: New file.
5792 * vec-perm-indices.h: Include int-vector-builder.h.
5793 (vec_perm_indices): Redefine as an int_vector_builder.
5794 (auto_vec_perm_indices): Delete.
5795 (vec_perm_builder): Redefine as a stand-alone class.
5796 (vec_perm_indices::vec_perm_indices): New function.
5797 (vec_perm_indices::clamp): Likewise.
5798 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5799 (vec_perm_indices::new_vector): New function.
5800 (vec_perm_indices::new_expanded_vector): Update for new
5801 vec_perm_indices class.
5802 (vec_perm_indices::rotate_inputs): New function.
5803 (vec_perm_indices::all_in_range_p): Operate directly on the
5804 encoded form, without computing elided elements.
5805 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5806 encoding. Update for new vec_perm_indices class.
5807 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5808 the given vec_perm_builder.
5809 (expand_vec_perm_var): Update vec_perm_builder constructor.
5810 (expand_mult_highpart): Use vec_perm_builder instead of
5811 auto_vec_perm_indices.
5812 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5813 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5814 or double series encoding as appropriate.
5815 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5816 vec_perm_indices instead of auto_vec_perm_indices.
5817 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5818 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5819 (vect_permute_store_chain): Likewise.
5820 (vect_grouped_load_supported): Likewise.
5821 (vect_permute_load_chain): Likewise.
5822 (vect_shift_permute_load_chain): Likewise.
5823 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5824 (vect_transform_slp_perm_load): Likewise.
5825 (vect_schedule_slp_instance): Likewise.
5826 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5827 (vectorizable_mask_load_store): Likewise.
5828 (vectorizable_bswap): Likewise.
5829 (vectorizable_store): Likewise.
5830 (vectorizable_load): Likewise.
5831 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5832 vec_perm_indices instead of auto_vec_perm_indices. Use
5833 tree_to_vec_perm_builder to read the vector from a tree.
5834 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5835 vec_perm_builder instead of a vec_perm_indices.
5836 (have_whole_vector_shift): Use vec_perm_builder and
5837 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5838 truncation to calc_vec_perm_mask_for_shift.
5839 (vect_create_epilog_for_reduction): Likewise.
5840 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5841 from auto_vec_perm_indices to vec_perm_indices.
5842 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5843 instead of changing individual elements.
5844 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5845 the vector in d.perm.
5846 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5847 from auto_vec_perm_indices to vec_perm_indices.
5848 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5849 instead of changing individual elements.
5850 (arm_vectorize_vec_perm_const): Use new_vector to install
5851 the vector in d.perm.
5852 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5853 Update vec_perm_builder constructor.
5854 (rs6000_expand_interleave): Likewise.
5855 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5856 (rs6000_expand_interleave): Likewise.
5858 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5860 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5861 to qimode could truncate the indices.
5862 * optabs.c (expand_vec_perm_var): Likewise.
5864 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5866 * Makefile.in (OBJS): Add vec-perm-indices.o.
5867 * vec-perm-indices.h: New file.
5868 * vec-perm-indices.c: Likewise.
5869 * target.h (vec_perm_indices): Replace with a forward class
5871 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5872 * optabs.h: Include vec-perm-indices.h.
5873 (expand_vec_perm): Delete.
5874 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5875 (expand_vec_perm_const): Declare.
5876 * target.def (vec_perm_const_ok): Replace with...
5877 (vec_perm_const): ...this new hook.
5878 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5879 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5880 * doc/tm.texi: Regenerate.
5881 * optabs.def (vec_perm_const): Delete.
5882 * doc/md.texi (vec_perm_const): Likewise.
5883 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5884 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5885 expand_vec_perm for constant permutation vectors. Assert that
5886 the mode of variable permutation vectors is the integer equivalent
5887 of the mode that is being permuted.
5888 * optabs-query.h (selector_fits_mode_p): Declare.
5889 * optabs-query.c: Include vec-perm-indices.h.
5890 (selector_fits_mode_p): New function.
5891 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5892 is defined, instead of checking whether the vec_perm_const_optab
5893 exists. Use targetm.vectorize.vec_perm_const instead of
5894 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5895 fit in the vector mode before using a variable permute.
5896 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5897 vec_perm_indices instead of an rtx.
5898 (expand_vec_perm): Replace with...
5899 (expand_vec_perm_const): ...this new function. Take the selector
5900 as a vec_perm_indices rather than an rtx. Also take the mode of
5901 the selector. Update call to shift_amt_for_vec_perm_mask.
5902 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5903 Use vec_perm_indices::new_expanded_vector to expand the original
5904 selector into bytes. Check whether the indices fit in the vector
5905 mode before using a variable permute.
5906 (expand_vec_perm_var): Make global.
5907 (expand_mult_highpart): Use expand_vec_perm_const.
5908 * fold-const.c: Includes vec-perm-indices.h.
5909 * tree-ssa-forwprop.c: Likewise.
5910 * tree-vect-data-refs.c: Likewise.
5911 * tree-vect-generic.c: Likewise.
5912 * tree-vect-loop.c: Likewise.
5913 * tree-vect-slp.c: Likewise.
5914 * tree-vect-stmts.c: Likewise.
5915 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5917 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5918 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5919 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5920 (aarch64_vectorize_vec_perm_const): ...this new function.
5921 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5922 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5923 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5924 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5925 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5926 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5927 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5929 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5930 check for NEON modes.
5931 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5932 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5933 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5934 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5936 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5937 the old VEC_PERM_CONST conditions.
5938 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5939 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5940 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5941 (ia64_vectorize_vec_perm_const_ok): Merge into...
5942 (ia64_vectorize_vec_perm_const): ...this new function.
5943 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5944 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5945 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5946 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5947 * config/mips/mips.c (mips_expand_vec_perm_const)
5948 (mips_vectorize_vec_perm_const_ok): Merge into...
5949 (mips_vectorize_vec_perm_const): ...this new function.
5950 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5951 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5952 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5953 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5954 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5955 (rs6000_expand_vec_perm_const): Delete.
5956 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5958 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5959 (altivec_expand_vec_perm_const_le): Take each operand individually.
5960 Operate on constant selectors rather than rtxes.
5961 (altivec_expand_vec_perm_const): Likewise. Update call to
5962 altivec_expand_vec_perm_const_le.
5963 (rs6000_expand_vec_perm_const): Delete.
5964 (rs6000_vectorize_vec_perm_const_ok): Delete.
5965 (rs6000_vectorize_vec_perm_const): New function.
5966 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5967 an element count and rtx array.
5968 (rs6000_expand_extract_even): Update call accordingly.
5969 (rs6000_expand_interleave): Likewise.
5970 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5971 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5972 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5973 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5974 (rs6000_expand_vec_perm_const): Delete.
5975 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5976 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5977 (altivec_expand_vec_perm_const_le): Take each operand individually.
5978 Operate on constant selectors rather than rtxes.
5979 (altivec_expand_vec_perm_const): Likewise. Update call to
5980 altivec_expand_vec_perm_const_le.
5981 (rs6000_expand_vec_perm_const): Delete.
5982 (rs6000_vectorize_vec_perm_const_ok): Delete.
5983 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5984 reference to the SPE evmerge intructions.
5985 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5986 an element count and rtx array.
5987 (rs6000_expand_extract_even): Update call accordingly.
5988 (rs6000_expand_interleave): Likewise.
5989 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5990 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5992 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5994 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5996 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5997 vector mode and that that mode matches the mode of the data
5999 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6000 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6001 directly using expand_vec_perm_1 when forcing selectors into
6003 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6005 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6007 * optabs-query.h (can_vec_perm_p): Delete.
6008 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6009 * optabs-query.c (can_vec_perm_p): Split into...
6010 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6011 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6012 particular selector is valid.
6013 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6014 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6015 (vect_grouped_load_supported): Likewise.
6016 (vect_shift_permute_load_chain): Likewise.
6017 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6018 (vect_transform_slp_perm_load): Likewise.
6019 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6020 (vectorizable_bswap): Likewise.
6021 (vect_gen_perm_mask_checked): Likewise.
6022 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6023 implementations of variable permutation vectors into account
6024 when deciding which selector to use.
6025 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6026 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6027 with a false third argument.
6028 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6029 to test whether the constant selector is valid and can_vec_perm_var_p
6030 to test whether a variable selector is valid.
6032 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6034 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6035 * optabs-query.c (can_vec_perm_p): Likewise.
6036 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6037 instead of vec_perm_indices.
6038 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6039 (vect_gen_perm_mask_checked): Likewise,
6040 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6041 (vect_gen_perm_mask_checked): Likewise,
6043 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6045 * optabs-query.h (qimode_for_vec_perm): Declare.
6046 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6047 (qimode_for_vec_perm): ...this new function.
6048 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6050 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6052 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6053 does not have a conditional at the top.
6055 2018-01-02 Richard Biener <rguenther@suse.de>
6057 * ipa-inline.c (big_speedup_p): Fix expression.
6059 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6062 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6065 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6069 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6070 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6071 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6072 cond_taken_branch_cost 3->4.
6074 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6076 PR tree-optimization/83581
6077 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6078 TODO_cleanup_cfg if any changes have been made.
6081 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6082 convert_modes if target mode has the right side, but different mode
6086 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6087 last argument when extracting from CONCAT. If either from_real or
6088 from_imag is NULL, use expansion through memory. If result is not
6089 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6090 the parts directly to inner mode, if even that fails, use expansion
6094 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6095 check for bswap in mode rather than HImode and use that in expand_unop
6098 Copyright (C) 2018 Free Software Foundation, Inc.
6100 Copying and distribution of this file, with or without modification,
6101 are permitted in any medium without royalty provided the copyright
6102 notice and this notice are preserved.